Potencia tu agente
El comando propio del publicador. No es un endpoint de PlugMCP.
Host a copy with our keys and logs.
npx -y @last9/[email protected]Claude Code no tiene enlace de instalación para un comando local. Copia el comando de abajo y ejecútalo en una terminal.
claude mcp add last9-mcp-server -- npx -y @last9/mcp-server@0.23.0Ejecútalo en una terminal.
{ "mcpServers": { "last9-mcp-server": { "command": "npx", "args": [ "-y", "@last9/[email protected]" ] } }}claude_desktop_config.json
¿URL remota o un comando en tu máquina? En qué se diferencian las dos instalaciones
Ejecutar en PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirDel publicador
Your AI agent doesn't know what's broken in production. This fixes that.
Last9 MCP Server connects Claude, Cursor, Windsurf, and any other MCP-capable AI assistant directly to your production observability data — logs, metrics, traces, exceptions, database queries, alerts, and deployments. The agent stops guessing and starts reading the actual signal.
Last9 MCP Server - Model Context Protocol server implementation for Last9
No binary to install. No tokens to manage. One URL, OAuth in your browser, done.
Find your org slug in your Last9 URL: app.last9.io/<org_slug>/...
claude mcp add --transport http last9 https://app.last9.io/api/v4/organizations/<org_slug>/mcp
Type /mcp, select last9, authenticate. That's it.
Settings > MCP > Add New MCP Server:
{
"mcpServers": {
"last9": {
"type": "http",
"url": "https://app.last9.io/api/v4/organizations/<org_slug>/mcp"
}
}
}
Click Connect, complete OAuth.
Requires v1.99+. Open Command Palette → MCP: Add Server, paste the URL, authenticate.
Or directly in settings.json:
{
"mcp": {
"servers": {
"last9": {
"type": "http",
"url": "https://app.last9.io/api/v4/organizations/<org_slug>/mcp"
}
}
}
}
Settings > Cascade > Open MCP Marketplace > gear icon (mcp_config.json):
{
"mcpServers": {
"last9": {
"serverUrl": "https://app.last9.io/api/v4/organizations/<org_slug>/mcp"
}
}
}
Settings > Connectors > Add custom connector. Name it last9, paste the URL, authenticate.
Requires admin access to your Claude organization.
Use this when your MCP client doesn't support HTTP transport, or when you need the server running locally.
Homebrew:
brew install last9/tap/last9-mcp
NPM:
Requires Node.js 22.23.4+, 24.22.1+ or 26.12.1+: earlier releases accept TLS certificates from the compromised eMudhra intermediates when the installer downloads the binary.
npm install -g @last9/mcp-server@latest
# or directly:
npx -y @last9/mcp-server@latest
Binary releases (Windows / manual):
Download from GitHub Releases:
Only admins can create tokens.
Homebrew:
{
"mcpServers": {
"last9": {
"command": "/opt/homebrew/bin/last9-mcp",
"env": {
"LAST9_REFRESH_TOKEN": "<your_refresh_token>"
}
}
}
}
NPM:
{
"mcpServers": {
"last9": {
"command": "npx",
"args": ["-y", "@last9/mcp-server@latest"],
"env": {
"LAST9_REFRESH_TOKEN": "<your_refresh_token>"
}
}
}
}
Where to paste this:
{
"mcp": {
"servers": {
"last9": {
"type": "stdio",
"command": "/opt/homebrew/bin/last9-mcp",
"env": {
"LAST9_REFRESH_TOKEN": "<your_refresh_token>"
}
}
}
}
}
For NPM: use "command": "npx" and add "args": ["-y", "@last9/mcp-server@latest"].
After downloading from GitHub Releases, extract and point to the full path:
{
"mcpServers": {
"last9": {
"command": "C:\\Users\\<user>\\AppData\\Local\\Programs\\last9-mcp-server.exe",
"env": {
"LAST9_REFRESH_TOKEN": "<your_refresh_token>"
}
}
}
}
The NPM route is easier on Windows — no path management.
--read-only (or LAST9_MCP_READ_ONLY=true) serves only tools annotated readOnlyHint: true, so no tool can create, change or delete alert rules, notification channels, dashboards or drop rules, acknowledge or reassign alerts, or mute groups. A tool without annotations counts as a write tool and is hidden.
To keep a few write tools, list them with --allow-write-tools (or LAST9_MCP_ALLOW_WRITE_TOOLS):
last9-mcp --read-only --allow-write-tools=acknowledge_alert,mute_alert_group
--allow-write-tools without --read-only is a startup error, because every write tool is already served.--toolsets: a tool is served only when it passes both. --toolsets=alerts --read-only --allow-write-tools=mute_alert_group serves the alerts read tools plus mute_alert_group; an allowed write tool outside the chosen toolsets stays hidden.--http mode and dump-tools honor the same flags and env vars: go run . dump-tools --read-only prints the read-only surface, with the server instructions it serves in instructions.create_alert creates a disabled rule that patch_alert enables), so allow such pairs together.get_workspace_overview — Recommended first call when a request names no service or tool (one that does goes straight to its tool): environments, which signals (logs, traces, metrics) were seen in the last hour, service and alert-group counts, firing alerts, and the current time in one cheap call. A signal not seen in that hour can still have older data, so it stays queryable. A failed sub-call leaves its field null and adds an errors[] entry. Served under every toolsetget_current_time — Current time, plus absolute bounds for last 2h, yesterday or since yesterday 9am IST, in any IANA zone or common abbreviation. Served under every toolsetget_service_summary — Ranked fleet (service, env) rows: interval request_count, throughput_rpm, HTTP 4xx/5xx counts, and gRPC error countsget_service_environments — Available environments for your services. Run this first — other APM tools need env from hereget_service_performance_details — Full breakdown: throughput, error rate, p50/p90/p95/avg/max, apdex, availability (summarized by default; full series with include_series)get_service_operations_summary — Operations grouped by HTTP endpoints, DB calls, messaging, HTTP clientsget_service_dependency_graph — Dependency map with throughput, latency, and error rates for upstream/downstream/infraget_apm_service_deviations — Compare a current window against an equal-duration baseline: regressions/improvements, Apdex reconciliation, and a terminal outcome (fleet or single service)get_exceptions — Server-side exceptions with service and span filtersFour tools that go directly at your database performance, derived from OpenTelemetry trace spans and, where traces are absent, infrastructure metrics such as CloudWatch. No extra instrumentation needed if you're already using OTel.
get_databases — Discover all databases across your infrastructure: DB type, host, throughput (queries/min), p95 latency, error rate, number of dependent services. Also discovers databases from infrastructure metrics such as CloudWatch, with no trace instrumentation needed — those rows carry an activity value instead of trace metricsget_database_slow_queries — The actual slowest query executions, ordered by duration, with trace IDs for drilling into full tracesget_database_queries — Query patterns and aggregates: how often a query runs, average/p95 duration, error rateget_database_server_metrics — Server-side metrics from the DB host itself (CPU, connections, buffer hit rates — depends on your DB system)Supports PostgreSQL, MySQL, MongoDB, Redis, Aerospike, and anything else OTel traces with a db_system attribute — plus databases discovered from infrastructure metrics such as CloudWatch, whose rows carry an activity value instead of trace metrics.
prometheus_range_query — PromQL range queries over any metricprometheus_instant_query — Instant queries; use rollup functions like avg_over_time, sum_over_timeprometheus_label_values — Label values for a given seriesprometheus_labels — All labels available for a seriesPoint these at a different datasource/cluster than the default by setting LAST9_DATASOURCE, or per call with datasource. An omitted datasource uses the default; a unique near miss that the matcher considers safe (such as a case variant) uses its closest match. A missing, extra or transposed letter is not guaranteed to resolve automatically: even one candidate can require confirmation when the edit could affect a region or environment code. Looser or ambiguous matches return candidates and run no query (not an error); unknown names return an error. An additional content block names the datasource used whenever you did not name it exactly. get_metric_status, get_service_profile, and get_apm_service_deviations resolve datasource the same way.
get_logs — Full JSON pipeline log queries (aggregations, filters, field extraction)get_service_logs — Log lines for a service, filterable by severity and body content; summarized over 20 lines unless response_mode is rawget_log_patterns — A service's log lines in a window clustered into message templates, with count, first/last seen, severity mix and a sampleget_emerging_log_patterns — Log templates that are new, increased, decreased or disappeared against a baseline window, with counts from bothget_log_attributes — Global catalog of attributes in the log schema for a time windowget_log_attributes_for_pipeline — Log fields actually present for an in-progress pipeline (scoped discovery), each with its exact filter_fieldget_drop_rules — Log drop rules from Last9 Control Planeadd_drop_rule — Create a new drop rule to cut log volume at the sourceget_traces — JSON pipeline trace queries for broad searches and aggregationsget_service_traces — Traces by exact trace ID or service name. Use this when you have a trace ID — it's fasterget_trace_attributes — Global catalog of attributes in the trace schemaget_trace_attributes_for_pipeline — Attributes actually present for an in-progress pipeline (scoped discovery), each with its exact filter_fieldget_trace_attribute_values — Distinct values for a trace attribute, optionally scoped to a pipelineget_trace_attribute_deviations — Ranks attribute values that differ between two bounded span cohorts (slow vs fast, error vs non-error, or two time windows). Correlation, not causeget_trace_waterfall — One exact trace as a parent/child waterfall with interval-union self-time, slowest spans, and graph warningsget_change_events — Deployments, config changes, rollbacks. Correlate incidents with what changedanalyze_change_impact — Did a change break anything? Compares a service and its direct callers before and after a change: golden signals, APM and trace-attribute deviations, new firing alerts, and a verdict (regression likely / no significant change / insufficient data)get_alert_groups — Configured Compass alert groups with metadata labels, team, tier, and rule counts — including groups with zero rules and groups that are not firingget_alert_config — Alert rule configurations — searchable by name, severity, type, tagsget_alerts — Currently firing alerts within a time windowget_alert_rule_state — Historical firing state (1/0) per alert rule over a time range, grouped by rule_id. Filterable by alert group, rule name, label filters, and state.get_notification_channels — Configured notification channels (Slack, PagerDuty, email, etc.). scope=channels lists the channels themselves (ids for attaching); the default attached lists per-alert-group attachmentsrecommend_alert_config — Read-only design guidance (lifecycle, validation gates, responder context) before creating an alert rulevalidate_alert_expression — Dry-run an alert expression against an alert group's KPIs without saving anythingdescribe_alert_chart — Read-only: the alertable signals (key, unit) on a Discover chart (discover-service, discover-exceptions)create_alert — Create an alert rule (static, or adaptive via adaptive_std_cmp) in an existing alert group; always created disabled as a draft. Create once; refine with update_alert / patch_alertcreate_alert_from_chart — Create a static-threshold alert rule from a Discover chart in one call. Like create_alert, the rule is created disabled as a draftupdate_alert — Full replacement of a rule's evaluation config (may return a replacement rule ID)patch_alert — Partial change to a rule: enable/pause, mute_until, responder properties, weekly recurring_mute_schedule (or clear_recurring_mute_schedule)delete_alert — Permanently delete an alert rulemute_alert_group — Mute (or unmute) every rule in an alert group until a deadlineacknowledge_alert — Take ownership of a firing alert, or release itreassign_alert — Hand an already-acknowledged alert to another usercreate_notification_channel — Create a Slack, PagerDuty, Opsgenie, email, or generic webhook channelattach_notification_channel — Route an existing channel to an alert group for breach, threat, or bothdetach_notification_channel — Remove one channel-to-alert-group attachmentdelete_notification_channel — Permanently delete a channel that has no attachmentsThe alert write tools (create_alert, update_alert, patch_alert, delete_alert, create_alert_from_chart, mute_alert_group, acknowledge_alert, reassign_alert, create_notification_channel, attach_notification_channel, detach_notification_channel, delete_notification_channel) need a token with write scope; delete_alert, detach_notification_channel and delete_notification_channel need delete scope instead. They are exposed only in the alerts toolset.
Rules, channels and attachments these tools create are marked created_via: last9-mcp together with the MCP client's name and version (on rules, the _l9_created_via_ and _l9_created_via_client_ annotations, which patch_alert and update_alert keep and callers cannot set). get_entity_alert_rules and get_notification_channels show the marker, and delete_alert, detach_notification_channel and delete_notification_channel tell agents to clean up only marked resources unless the user names one.
list_dashboards — All custom dashboards in your org: IDs, names, and metadataget_dashboard — Full dashboard definition by ID, including panels and queriesvalidate_dashboard — Read-only lint + execute + classify for a saved dashboard id or an inline dashboard_definition over a ≤24h window. Never creates or updates dashboardscreate_dashboard — Create a net-new custom dashboard once (panels, queries, metadata). After the id is returned, edit panels with patch_dashboard.update_dashboard — Restructure most of an existing dashboard by ID (full replacement; readonly system dashboards return an error). For individual panel edits use patch_dashboardpatch_dashboard — Targeted edits to an existing dashboard by ID (add/update/remove/move panel, set title, set variable); returns a compact diff, not the full dashboarddelete_dashboard — Delete a custom dashboard by IDlist_dashboard_snapshots — Frozen point-in-time snapshots for a dashboard (metadata only)get_dashboard_snapshot — Full frozen snapshot including panel data for RCA / shareable viewsdelete_dashboard_snapshot — Delete a frozen snapshot by IDRequires continuous profiling enabled for the org. Discover services first with get_profile_services, then pull a flamegraph or ranked functions.
get_profile_services — Services that have profiling data in the window (index before querying)get_flamegraph — Nested flamegraph tree for one service (cpu default; also alloc, wall)get_top_functions — Self-sample ranking of hottest functions for one serviceget_profile_summary — Short natural-language triage of the profile for one serviceRead-only tools against the org's Grafana instance (via Last9's Grafana proxy). Credential fields are never returned to the model. Enable with LAST9_TOOLSETS=grafana (or leave toolsets unset for all tools).
grafana_search_dashboards — Search dashboards by title substring (paginated; truncated: true when the cap is hit)grafana_get_dashboard — Dashboard summary by uid (panels, variables, PromQL targets); full_json=true for raw Grafana JSONgrafana_list_folders — Folder treegrafana_list_folder_dashboards — Dashboards in one folder (paginated)grafana_list_datasources — Datasource inventory without credentialsdid_you_mean — When the agent isn't sure about an entity name, this returns the closest matches from your catalog (services, environments, hosts, databases, K8s deployments/namespaces, jobs). Up to 3 suggestions with similarity scores. The server calls this automatically before most tools when a name lookup returns empty.get_service_profile — What a service's telemetry actually looks like, before you query it: which signals exist, language and runtime, deployment environments, the shape of its logs, and a recommended ingest fix where one applies. Lets the agent skip trace tools when a service has no traces, and parse severity from the log body when SeverityText is empty instead of filtering on it and finding nothing.Deep links on every response. Every tool returns a deep_link field — a direct URL into the Last9 dashboard for that exact query and time range. The agent can hand you the link; you click it; you're there.
Toolsets. By default the server exposes every tool. Automation hosts that only need investigation (logs/traces/metrics/profiles) can set LAST9_TOOLSETS=investigate (or pass --toolsets=investigate) so tools/list stays small without client-side mass-disable. Named packs: logs, traces, metrics, alerts, dashboards, profiles, grafana, investigate, all. Unknown names fail fast. get_current_time and get_workspace_overview are served under every toolset. In HTTP mode a request can narrow this further with the X-Last9-Toolsets header (see Per-request toolsets). The metrics pack alone does not include list_datasources or did_you_mean — use investigate (or combine toolsets) when you need those discovery helpers.
Tool reference resources. Long logjson/tracejson/service-logs/metrics manuals are MCP resources (last9://reference/logjson, last9://reference/tracejson, last9://reference/service_logs, last9://reference/metrics, last9://reference/investigation), not always-on tool description text. Critical query rules stay on the tool description so agents that never call resources/read still get correct construction guidance. Discover org-specific fields with get_log_attributes / get_log_attributes_for_pipeline (and the trace equivalents)—they are not injected into descriptions.
Chunked large results. get_logs and get_traces handle large result sets through chunking rather than truncating. The default limit is 5000 entries for logs; configurable via LAST9_MAX_GET_LOGS_ENTRIES.
export LAST9_REFRESH_TOKEN="your_refresh_token"
export LAST9_HTTP=true
export LAST9_PORT=8080
./last9-mcp-server
Server starts at http://localhost:8080/mcp.
The Streamable HTTP handler runs in stateless mode, so any request is served independently. An initialize handshake and an Mcp-Session-Id header are optional — clients that send them still work (the header is accepted and ignored), and clients can also skip straight to tools/list / tools/call. Every tool is an independent request/response query; the server issues no server→client notifications, so GET /mcp (the SSE stream) returns 405.
# List tools — a session handshake is optional in stateless mode
curl -s -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}'
# Call a tool
curl -s -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/call",
"params": {
"name": "get_service_logs",
"arguments": {
"service_name": "your-service-name",
"lookback_minutes": 30,
"limit": 10
}
}
}'
In HTTP mode a request can narrow its tool surface with the X-Last9-Toolsets header, or the toolsets query parameter for clients that cannot set headers (/mcp?toolsets=logs,traces). A non-blank header wins when both are sent. The value takes the same comma-separated names as --toolsets. The request sees the intersection with the startup --toolsets: it can narrow, never widen, and all or an empty value keeps the startup set. An unknown name returns 400 naming it. tools/list, tools/call and the initialize instructions all follow the effective set; get_current_time and get_workspace_overview stay in every set.
curl -s -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-H "X-Last9-Toolsets: logs,alerts" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}'
Each distinct effective set is served by its own server, built on its first request (tens of milliseconds) and kept until shutdown; a request restating the startup set reuses the startup server. There is no eviction: only named toolsets are accepted, so the cache holds at most one server per combination of them (135 today, about 75 MB of live heap together; a full cache settled near 230 MB resident in testing).
git clone https://github.com/last9/last9-mcp-server.git
cd last9-mcp-server
go build -o last9-mcp-server
LAST9_HTTP=true ./last9-mcp-server
LAST9_HTTP=true is for local development. For actual usage, the hosted HTTP endpoint is easier.
start_time_iso/end_time_iso, or time_iso) take precedence over lookback_minutes.lookback_minutes.2026-02-09T15:04:05Z.YYYY-MM-DD HH:MM:SS is accepted for compatibility only.since yesterday 9am IST to RFC3339 bounds with get_current_time.No parameters. Probes the last 60 minutes concurrently, each sub-call with a short timeout, and returns current_time, lookback_minutes, environments, telemetry (logs/traces/metrics booleans), service_count (distinct services get_service_summary ranks, every env), alert_group_count, firing_alert_count and errors ([{field, error}]). A field whose sub-call failed is null (unknown), never 0 or false. The probes are not narrowed by --toolsets: toolsets filter which tools are served, and the API token scopes what the server can read.
range (string, optional): last 2h, past 30 minutes, last 7d (s, m, h, d = 24h, w = 7d; at most 366d); today; yesterday; since <point>; <point> to <point>. A point is RFC3339, or today / yesterday / a weekday (monday, mon: its latest occurrence, today included) / YYYY-MM-DD with an optional 9am, 9:30pm or 14:30. A trailing zone overrides timezone for both bounds; a zone anywhere else is an error. Omit for just now.timezone (string, optional): IANA name or abbreviation (IST = Asia/Kolkata, PST/PDT, EST/EDT, UTC, GMT). Default UTC. Unknown zones are an error, and so is the ambiguous CST (pass America/Chicago or Asia/Shanghai).Returns now (and start/end with range) as {iso, epoch_seconds, local}; local appears only for a non-UTC zone.
limit (integer, optional): Max exceptions. Default: 20.lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional): Absolute time range.service_name (string, optional): Filter by service.span_name (string, optional): Filter by span name.env (string, optional): Filter by environment.start_time_iso / end_time_iso (string, optional)env (string, optional): PromQL regex. Defaults to .*. Exact match needs anchors (e.g. ^prod$).sort_by (string, optional): request_count (default), throughput_rpm, http_4xx_count, http_5xx_count, or grpc_error_count.limit (integer, optional): Max ranked rows. Omit or 0 means 10; values above 100 clamp to 100.start_time_iso / end_time_iso (string, optional)All other APM tools require an
envvalue. Use""if this returns empty.
Returns a per-series summary by default: points, missing_points (availability buckets with no traffic, excluded from the stats), min, p50, p90, p99 (nearest-rank), max, mean, last, baseline_mean (mean of the first half of the valid points, not of the window), delta_pct (second half of the valid points vs baseline; absent if baseline is 0) and up to 3 change_points (timestamp, before_mean, after_mean), for up to 4 series per metric and every response-time quantile (highest mean first; series_omitted counts the rest). A series with points: 0 carries no statistics, and baseline_mean/delta_pct need at least 6 points (3 per half). full_series marks that raw points are available via include_series.
service_name (string, required)lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)env (string, optional): Defaults to prod.top_n (integer, optional): Max entries per top-operations/top-errors list. Default: 5, or 10 with include_series=true. Max 100.include_series (boolean, optional): Return every point of every series instead of the summary. Default: false.service_name (string, required)lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)env (string, optional): Defaults to prod.service_name (string, optional)lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)env (string, optional): Defaults to prod.service_name (string, optional): Omit for fleet scope; provide for one service and its operation correlations.lookback_minutes (integer, optional): Current window. Default: 60.start_time_iso / end_time_iso (string, optional): Explicit current window.baseline_start_time_iso / baseline_end_time_iso (string, optional): Explicit baseline. Defaults to the immediately preceding equal-duration window.datasource (string, optional): Omit for the default; approximate names resolve as for the Prometheus tools. The response names the default or closest match used; ambiguous names return candidates without querying.env (string, optional): Defaults to prod.max_services / max_operations (integer, optional): Default 10, max 10 each.env (string, optional): Filter by environment. Accepts a regular expression. Default: all.lookback_minutes (integer, optional): Default: 60. Window may not exceed 7 days.start_time_iso / end_time_iso (string, optional)db_system (string, optional): e.g. postgresql, mysql, mongodb, redis.host (string, optional): Database host (net_peer_name).service_name (string, optional): Calling service name.env (string, optional)min_duration_ms (float, optional): Minimum query duration in ms.lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)limit (integer, optional): Default: 20.db_system (string, optional)host (string, optional)service_name (string, optional)env (string, optional)lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)limit (integer, optional): Default: 20.db_system (string, required): e.g. postgresql, mysql, mongodb, redis, aerospike.host (string, optional)lookback_minutes (integer, optional): Default: 60.start_time_iso / end_time_iso (string, optional)query (string, required): The PromQL query.start_time_iso / end_time_iso (string, optional): Defaults to last 60 min.lookback_minutes (float, optional): Default: 60.query (string, required)time_iso (string, optional): Defaults to now.lookback_minutes (float, optional)match_query (string, optional): PromQL filter.label (string, required): Label name.lookback_minutes (float, optional): Default: 60.start_time_iso / end_time_iso (string, optional)When no window is given and match_query names one metric (metric{...} or {__name__="metric"}), an empty 60-minute result is retried once over 4 days; a second text block says which window the result covers.
match_query (string, optional): PromQL filter.lookback_minutes (float, optional): Default: 60.start_time_iso / end_time_iso (string, optional)Retries an empty single-metric lookup over 4 days, like prometheus_label_values.
logjson_query (array, required): JSON pipeline query.lookback_minutes (integer, optional): Default: 5.start_time_iso / end_time_iso (string, optional)limit (integer, optional): Server default: 5000.index (string, optional): physical_index:<name> or rehydration_index:<block_name>.response_mode (string, optional): summary (default; also for null) or raw. A raw-line result over 20 lines comes back as summary: lines_summarized, severity_counts (top six labels, the rest in other_severities), first_timestamp/last_timestamp, a 10-bucket histogram over the query window (over the returned lines, with truncated: true, when the limit cut the result), and the top patterns (template, count, up to two samples) with the rest totalled in other, at most 4000 characters; top-level metadata that still does not fit is dropped and named in dropped_fields. raw returns the lines. Aggregate results are never summarized.For log-based service inventory, query physical_index_service_count first:
sum by (name, service_name, env) (physical_index_service_count{destination="logs"})
Use service_name as ServiceName, env as the environment when present, and name as the physical index name. If name="default", omit index; for a non-default physical index selected by the user, pass index: "physical_index:<name>". If the backend rejects explicit physical index filtering, retry without index and report that explicit physical index filtering is unavailable for that backend.
service_name (string, required)lookback_minutes (integer, optional): Default: 60.limit (integer, optional): Default: 100, or 20 with response_mode: "raw".env (string, optional)severity_filters (array, optional): e.g. ["error", "warn"]. OR logic.body_filters (array, optional): e.g. ["timeout", "failed"]. OR logic.start_time_iso / end_time_iso (string, optional)index (string, optional)response_mode (string, optional): summary (default; also for null) or raw. Over 20 lines, logs is replaced by the same summary as get_logs, with the other fields kept; raw returns the entries.Multiple filter types combine with AND. Each array uses OR internally.
Use get_logs for broad aggregate counts first; use get_service_logs only after narrowing to a service/env/index and a small sample set.
service_name (string, required)env (string, optional)start_time_iso / end_time_iso (string, optional)lookback_minutes (integer, optional): Default: 60.severity_filters / body_filters (array, optional): OR logic, as in get_service_logs. Set severity_filters only when get_service_profile reports severity_set all; with none or partial it drops lines whose severity is in the body.index (string, optional)sample_size (integer, optional): Lines sampled from the window, newest first. Default: 1000, maximum: 5000.top_n (integer, optional): Templates returned by count. Default: 20, maximum: 100. The rest are totalled in other.Clustering runs in the MCP server over the sampled lines, so counts, first_seen and last_seen are within the sample; window.sample_limit_reached says when the sample may miss matching lines (it holds sample_size lines and the API does not report an exact total_matching_lines equal to that, or the search was logs_truncated), in which case it holds only the window's newest lines. Each template and sample is cut at 500 characters, ending in …, so a body that does not cluster (such as JSON without a message field) cannot fill the response. A search with failed chunks is a tool error; a truncated or capped search still returns patterns, flagged by sample_limit_reached. An empty patterns list means nothing matched only while sample_limit_reached is false; when it is true the search stopped early and may have missed matching lines.
Takes every get_log_patterns parameter (the time range is the current window), plus:
baseline_start_time_iso / baseline_end_time_iso (string, optional): Set both or neither. Default: the window of equal length immediately before the current one.min_count (integer, optional): Lines a changed pattern needs in the window it is larger in: current for new/increased, baseline for decreased/disappeared. Default: 3 (or sample_size if smaller), maximum: sample_size.change_factor (number, optional): Frequency ratio, greater than 1, that a pattern seen in both windows must reach (increased) or fall to the inverse of (decreased). Default: 2.Both thresholds must hold, so a high-volume template whose frequency did not change by change_factor is never reported. Returns only changed templates (new, increased, decreased, disappeared) with current and baseline counts; unchanged_patterns counts templates in both windows that changed by less than change_factor, and below_min_count_patterns counts changes dropped for having fewer than min_count lines. Frequencies are compared per minute while both samples hold every matching line, per minute estimated from a capped sample's exact total_matching_lines, and otherwise as shares of the sample, where increased/decreased describe composition rather than rate (comparison_basis). An empty patterns list means no change above both thresholds, not unchanged logs: rarer changes are counted in below_min_count_patterns, and a window with sample_limit_reached compares samples, not whole windows.
lookback_minutes (integer, optional): Default: 15.start_time_iso / end_time_iso (string, optional)region (string, optional)index (string, optional)pipeline (array, required): Prior filter stages to scope discovery, e.g. [{"type":"filter","query":{"$eq":["ServiceName","<service>"]}}].lookback_minutes (integer, optional): Default: 15.start_time_iso / end_time_iso (string, optional)region (string, optional)index (string, optional)No parameters. Lists drop rules via GET /otel_settings/drop?region=....
name (string, required)filters (array, required): Each filter: key, value, operator (equals/not_equals), conjunction (and).attributes["key_name"] or resource.attributes["key_name"] (required by the Last9 API).POST /otel_settings/drop?region=...&cluster_id=....Use for broad searches and aggregations. For exact trace ID lookup, use get_service_traces.
tracejson_query (array, required)start_time_iso / end_time_iso (string, optional)lookback_minutes (integer, optional): Default: 60.limit (integer, optional): Default: 5000.Exactly one of trace_id or service_name is required.
trace_id (string, optional): Default lookback: 72 hours.service_name (string, optional): Default lookback: 60 min.lookback_minutes (integer, optional)start_time_iso / end_time_iso (string, optional)limit (integer, optional): Default: 10.env (string, optional)lookback_minutes (integer, optional): Default: 15.start_time_iso / end_time_iso (string, optional)region (string, optional)pipeline (array, required): Prior filter stages to scope discovery, e.g. [{"type":"filter","query":{"$eq":["ServiceName","<service>"]}}].lookback_minutes (integer, optional): Default: 15.start_time_iso / end_time_iso (string, optional)region (string, optional)tag_name (string, required): Attribute name from get_trace_attributes (e.g. resource_department or attributes['http.method']).pipeline (array, optional): Prior filter stages to scope the values; omit for global values.lookback_minutes (integer, optional): Default: 15.start_time_iso / end_time_iso (string, optional): Historical RFC3339 bounds; take precedence over lookback_minutes.region (string, optional)comparison_mode (string, required): latency, errors, or time.service_name (string, required)environment (string, required): Exact deployment.environment value.operation (string, optional)filters (array, optional): Trace JSON filter conditions.candidate_attributes (array, optional): Maximum 8; omit for bounded discovery.latency_threshold_ms (number, optional): Required for latency mode; rejected for other modes.start_time_iso / end_time_iso (string, optional)lookback_minutes (integer, optional): Default: 15. Maximum: 15.baseline_start_time_iso / baseline_end_time_iso (string, optional): Required for time mode; non-overlapping and equal in duration to the target window.minimum_cohort_size (integer, optional): Default: 100. Minimum: 20.minimum_value_support (integer, optional): Default: 20. Minimum: 10.limit (integer, optional): Default: 10. Maximum: 10.Requires the companion backend capability to be enabled.
trace_id (string, required)environment (string, optional)start_time_iso / end_time_iso (string, optional)lookback_minutes (integer, optional): Default: 4320 (72 hours).selected_span_id (string, optional): Returns attributes, events, and links for that span only.max_spans (integer, optional): Default: 500. Maximum: 1000.Returns an investigation-evidence/v1 envelope; the waterfall is under data.
start_time_iso / end_time_iso (string, optional)lookback_minutes (integer, optional): Default: 60.service_name (string, optional)env (string, optional)event_name (string, optional): Call without this first to get available_event_names.Pass exactly one of change_event or service + timestamp.
change_event (object, optional): One element of get_change_events change_events[], unchanged. Uses metric.service, metric.deployment_environment and the earliest values[].timestamp.service (string, optional): Required with timestamp when change_event is omitted.timestamp (string, optional): RFC3339 or unix seconds, rounded down to the minute.env (string, optional): Exact environment. Default: the event's, else all.baseline_minutes (integer, optional): Default: 60. Clamped to 30..120.after_minutes (integer, optional): Default: baseline_minutes. Clamped to 30..120 and capped at now.regression likely when any holds: error rate up at least 1 percentage point, p95 up at least 40%, or throughput (rpm) down at least 50% for the service or one of its top 10 callers, ranked by throughput over both windows together (only with at least 100 requests before the change and, but for throughput, after it; p95 counts only latency_env's requests); an alert for them starts firing after the change and is still firing at the end of the after window (matched case-insensitively by alert group name or a service, service_name or service.name label, else annotation (never free text such as a summary); with env, alerts labelled with another environment, ignoring case, are dropped and unlabelled ones are marked env_match: unlabelled); a regression from get_apm_service_deviations. insufficient data when the after window is under 5 minutes (including a change in the future), the service has too few requests, or golden signals, the direct-caller lookup, alerts or get_apm_service_deviations failed, wholly or in any query listed in deviations.apm.unavailable: missing evidence never yields no significant change, and a regression found anyway names what failed. no significant change is a valid answer, not a failure. Trace-attribute deviations run only when env is set, over at most 15 minutes each side; they explain a verdict but never set one, so skipping them or failing to fetch them leaves it unchanged. Each list keeps 10 items and reports how many it dropped; a failed upstream call leaves an error on its section only. Golden-signal latency comes from one environment in both windows, named in latency_env: with no env, the busiest. Golden-signal error rate counts 5xx and gRPC errors; deviations.apm also counts error-status spans and 429s. A caller's signals are its own server spans, so client-only callers show 0 rpm.
regressions lists one entry per trigger behind regression likely, in reason's order, and is [] for the other verdicts. Each has service, role (changed or dependent), env, metric (exactly one of error_rate_pct, latency_p95_ms, throughput_rpm, alert, apm_deviation) and baseline/after in the metric's unit (%, ms, rpm, or the APM signal's own value; null for alert). An alert item adds rule_id and state, the instance's state in the newest alerts-monitor slice of the after window that lists it, always firing: an alert that started after the change but resolved by the end of the after window is not a regression, so it files no ticket; it stays in alerts.started with its state, and reason counts it. env is the environment the metric was measured in: the requested env, else error rate and throughput sum every environment (env: "") while p95 uses the busiest, latency_env; an alert uses its own env label and an APM deviation its own env. Every regression is service-wide, so none names an endpoint.
Configured Compass alert-group inventory for changeboard / label-coverage audits. Includes groups with zero rules and groups that are not firing. Does not return PromQL.
alert_group_name / alert_group_type / data_source_name (string, optional): Case-insensitive substring match.team / tier (string, optional): Exact case-insensitive match on configured metadata.label_key + label_value (string, optional): Must be set together. Exact case-insensitive match on one metadata.labels pair — both key and value.Returns compact JSON {"count":N,"groups":[...]} with id, name, type, entity_class, team, tier, metadata.labels, and rule counts. Empty team / labels means unset.
search_term (string, optional): Free-text search across name, group, data source, tags.rule_name (string, optional)severity (string, optional)rule_type (string, optional): static or anomaly.alert_group_name / alert_group_type / data_source_name (string, optional)tags (array, optional): All must match (AND logic).time_iso (string, optional): Evaluation time in RFC3339.window (integer, optional): Lookback in seconds. Default: 900. Range: 60–86400.lookback_minutes (integer, optional): Range: 1–1440.start_time (integer, required): Unix epoch start of the range (inclusive).end_time (integer, required): Unix epoch end of the range (inclusive).step (integer, required): Resolution in seconds between samples. The number of samples ((end_time - start_time) / step + 1) is capped at 100.alert_group_id (string, optional): Filter by alert group ID.rule_name (string, optional): Regex filter on rule name.alert_group_name (string, optional): Regex filter on alert group name.label_filters (string, optional): Comma-separated key=value label filters.state (string, optional): Filter by state (e.g. firing).Returns a JSON map of rule_id -> [{timestamp, is_firing}]. A timestamp at which a rule is absent from the upstream response is reported as is_firing=0 — this means "not observed as firing", not a confirmed normal state.
scope (string, optional): attached (default) returns one row per alert-group attachment; the id is an attachment id, as taken by detach_notification_channel. channels returns the org's channels (id, name, type, send_resolved, created_via); the id is a channel id, as taken by attach_notification_channel. Destinations, headers, and certificates are never returned.Returns configured notification channels (Slack, PagerDuty, email, webhooks, etc.) as a TSV table. Attachment rows (service_fqid set) show global false and services -; services lists only legacy service bindings. On attachment rows send_resolved is the row's stored default, not what delivery uses; resolve notifications follow the channel's own send_resolved (scope=channels). Both scopes include created_via (last9-mcp for rows an MCP agent created, else -).
Read-only. Returns lifecycle guidance, validation gates, and responder-context requirements; it is not evidence by itself, so ground thresholds with telemetry tools.
signal (string, required): Metric or user-impact signal, including unit and dimensions.objective (string, required): Symptom or user impact to detect.rule_type (string, optional): static (default) or anomaly.severity (string, optional): breach (default) or threat.environment (string, optional): Environment and traffic profile.Dry-runs an expression without saving. Use before create_alert / update_alert.
entity_id (string, required): Alert group/entity UUID.expression (string, required): Arithmetic over indicators named by their KPI's exact name, e.g. errors / requests. At most 3 indicators. Write names with spaces, dots or dashes as-is (Test Alert / errors), exactly as create_alert saves them.indicators_used (object, required, non-empty): Indicator name to KPI UUID owned by entity_id. Use each KPI's exact name: this dry-run accepts any name, but create_alert and update_alert reject expression_args keys that are not the bound KPI's name.Only the arithmetic expression is checked (pass the inner metric expression, not an adaptive_std_cmp(...) wrapper); condition / alert_condition are not. The at-least-one-timeseries check runs only when more than one indicator is used.
Returns {"valid":true} or {"valid":false,"reason":"..."}; an invalid expression is a normal result, not a tool error.
Read-only. Lists the alertable signals (key, unit, eval_query, query_kind) on a covered Discover chart. Viewer-role tokens are rejected.
surface (string, required): discover-service or discover-exceptions; any other surface, including custom dashboards, is rejected before calling the API (use create_alert with a KPI instead).chart_key (string, required): e.g. apdex, error_rate, response_time, exception_count.service_name (string, required)env (string, optional)attributes (object, optional): Extra identity dimensions, using the key names this tool reports.Creates a rule in an existing alert group, always disabled (a draft, even if is_disabled=false is passed). Call recommend_alert_config first, then promote with patch_alert is_disabled=false only after the user confirms. Create once and keep the returned rule ID. A rule_name the alert group already uses (case-insensitive) is refused without creating anything, naming the existing rule's id.
entity_id (string, required): Alert group/entity UUID.alert_rule (object, required): Complete typed rule body, including primary_indicator and expression_args bound to real KPIs of the entity. Each expression_args key must be the exact name of the KPI it binds (not a placeholder such as x), and primary_indicator and expression use those names. Static rules require condition, alert_condition, and eval_window (0–60 minutes). Adaptive rules use expression = adaptive_std_cmp(<primary_indicator>, sensitivity, window); the three fields are then optional (defaults expr > 0, count_true(result) >= 1, 1). algorithm cannot be set. Responder context goes in properties (description, runbook, annotations, query_mode).Notification routing is configured on the alert group, not the rule.
Creates a static-threshold rule from a chart described by describe_alert_chart. Like create_alert, the rule is created disabled as a draft and does not fire; promote it with patch_alert is_disabled=false only after the user confirms. The chart API creates rules enabled, so the tool disables the rule right after creating it; if that fails, the result is an error naming the rule id, which may be live.
surface / chart_key / service_name / signal_key / name (string, required)env (string, optional)attributes (object, optional)threshold (number, optional): In the signal's unit; must be 0 or more. Score-unit signals such as apdex need an explicit value.threshold_operator (string, optional): > (default), <, >=, <=, ==, !=.eval_window (integer, optional): Minutes, 1–60. Default: 5.bad_minutes (integer, optional): 1–eval_window. Default: 3.severity (string, optional): breach or threat.description (string, optional)group_id (string, optional): Normally omit.Full replacement of a rule's evaluation config. Retrieve the rule first with get_alert_config or get_entity_alert_rules. Refuses without changing anything when the rule does not exist or belongs to another alert group. May soft-delete and recreate the rule, so use the returned replacement ID afterwards.
entity_id (string, required)id (string, required): Alert rule UUID.alert_rule (object, required): Complete configuration; same shape as create_alert.Partial change that preserves the rule ID. At least one change is required. Refuses without changing anything when the rule does not exist or belongs to another alert group.
entity_id (string, required)id (string, required): Alert rule UUID.is_disabled (boolean, optional): false promotes/enables, true pauses.mute_until (integer, optional): Unix epoch deadline; 0 unmutes.properties (object, optional): description, runbook, annotations, query_mode. Each field sent replaces the stored one; omitted or blank fields keep their stored value. annotations replaces the whole map, so send every key to keep; {} clears it.recurring_mute_schedule (object, optional): Replaces the weekly schedule: enabled (default false), timezone (IANA name, required), and windows (1–20 of weekdays as sun..sat, start_time / end_time as 24-hour HH:MM). An end at or before the start ends the next day. Send every window to keep.clear_recurring_mute_schedule (boolean, optional): Removes the schedule entirely; cannot be combined with recurring_mute_schedule. Differs from enabled=false, which keeps the windows stored.Viewer-role users may only change mute_until. Rule-level mute changes are rejected while the group or entity is muted. Read changes back with get_entity_alert_rules, which shows mute_until, recurring_mute_schedule and properties.
Permanently deletes a rule. Verify with get_alert_config and get explicit user confirmation first. Refuses without deleting anything when the rule does not exist or belongs to another alert group.
entity_id (string, required)id (string, required): Alert rule UUID.Mutes every current and future rule in a group; to mute one rule use patch_alert with mute_until. Requires editor access.
entity_id (string, required): Alert group/entity UUID.until (integer, required): Unix epoch seconds when the mute ends. 0 unmutes; a negative value such as -1 mutes indefinitely.Takes ownership of a firing alert, or releases it. Does not resolve or mute the alert. Acknowledging an alert you already own is a safe no-op. ack refuses without recording anything unless get_alerts lists that exact alert (all three ids) within the last hour.
alert_group_id / rule_id / rule_type (string, required): Copy from the alert in get_alerts.action (string, optional): ack (default) or unack. Only the owner or an org admin can unack.mute_until (integer, optional): ack only. Records a mute you already applied with mute_alert_group or patch_alert; does not mute anything.Hands an alert that already has an owner to another user; acknowledge it first if it has none. When nothing is reassigned because the alert has no owner, it says whether get_alerts lists that alert in the last hour at all. Last9 does not notify the new owner.
alert_group_id / rule_id / rule_type (string, required): Copy from the alert in get_alerts.owner_email (string, optional): Preferred; matched case-insensitively against active org users.owner_user_id (string, optional): Provide exactly one of owner_email or owner_user_id.Creates a channel; it routes nothing until attached with attach_notification_channel. Create once and keep the id; there is no update tool. Returns only id, name, type, and send_resolved.
name (string, required): Unique per type; no colon.type (string, required): slack, pagerduty, opsgenie, email, or generic_webhook.destination (string, required): Slack Incoming Webhook URL (or channel ID with slack_app_mode), PagerDuty routing key, Opsgenie API key, comma-separated work email addresses on any domain, not only the org's (the API rejects personal-provider addresses such as Gmail, and an org on a personal login accepts only its admin's address), or an HTTPS webhook URL.slack_app_mode (boolean, optional): Slack only. Default: false.mention_handles (string, optional): Comma-separated, e.g. user|U123ABC|name, special|here, special|channel.webhook_headers (object, optional): generic_webhook only.webhook_ca_certificate (string, optional): generic_webhook only; one PEM CA certificate up to 64 KiB.send_resolved (boolean, optional): Default: true.Routes an existing channel to one alert group. Idempotent. Attachments copy the channel's delivery settings (headers, certificate, mentions) at attach time; if the channel's settings change later, reattach.
channel_id (integer, required): A channel id (get_notification_channels with scope=channels), not an attachment id.alert_group_id (string, required): From get_alert_groups.severity (string, required): breach, threat, or both. both makes two attachments; if the second fails, the first stays attached and the error names the severity to retry.Removes one attachment. Requires a delete-scope token; a write-scope token cannot detach.
attachment_id (integer, required): The id from get_notification_channels with scope=attached. Channel ids are refused.alert_group_id (string, optional): Refuses if the attachment belongs to a different group.Permanently deletes a channel that has no attachments; detach its attachments first. Requires a delete-scope token. Confirms the deletion by re-listing channels.
channel_id (integer, required): The id from get_notification_channels with scope=channels. Attachment ids are refused.query (string, required): The name to search for — partial, misspelled, or abbreviated.type (string, optional): Restrict to entity type: service, environment, host, database, k8s_deployment, k8s_namespace, job.Returns up to 3 closest matches with similarity scores. Use this before any tool call where the entity name is uncertain. If a previous call returned empty results, try this before retrying.
service_name (string, required): Service to derive a telemetry profile for.datasource (string, optional): Datasource name. Omit for the default; an approximate name resolves as for the Prometheus tools.Returns a short investigation brief followed by the full profile as raw JSON: signal presence (logs/traces/metrics as present, absent, or unknown), language and runtime, deployment environments, log signal_shape (log_format, severity_set, level_field, trace_context_in_logs), and a recommended ingest fix where one applies. When the backend reports freshness, the brief adds a last seen line with freshness.logs.last_seen and freshness.traces.last_seen (RFC3339, or none when null), and a traces stale warning when freshness.traces_stale is true: spans stopped while logs continued, so investigate a regression around freshness.traces.last_seen, not missing setup. Derived upstream and cached with a ~15 minute TTL.
Call it before any service-scoped investigation so tool selection matches the service's actual telemetry — skip trace tools when traces is absent, and when severity_set is none or partial parse severity from level_field in the log body rather than using severity_filters. metrics is always unknown and dependencies is unpopulated in v1. When logs and traces are both absent, confirm the name with did_you_mean before concluding the service is unmonitored.
No parameters. Returns all custom dashboards in the org as a JSON array with id, name, and metadata.
id (string, required): Dashboard UUID.region (string, optional): Region for panel query population. Defaults to configured datasource region.Read-only. Never creates or updates dashboards. Accept exactly one of dashboard_id or dashboard_definition.
dashboard_id (string, optional): Saved dashboard UUID to validate.dashboard_definition (object, optional): Inline unsaved dashboard body (true dry run).start_time_iso / end_time_iso (string, optional): Validation window (RFC3339). Must be ≤ 24h.region (string, optional): Region for panel query execution.Returns dashboard_validation/v1: per-panel lint + execute classification (data / no_data / invalid / error). Day-1 empty results classify as valid_no_data without diagnose probes.
Net-new only. After this call returns dashboard.id, edit panels with patch_dashboard (or restructure with update_dashboard) — do not create again to add, trim, or fix panels.
dashboard (object, required): Dashboard definition with name and panels[]. Each panel requires name, version, layout (x, y, w, h), visualization.type, and queries[].metadata (object, optional): Dashboard metadata — _category and _type fields (e.g. {"_category":"custom","_type":"metrics"}).Use only to restructure most of a dashboard; for adding, editing, removing or moving individual panels use patch_dashboard. Full replacement by id (same body as create): every panel must be re-sent exactly.
id (string, required): Dashboard UUID to update.dashboard (object, required): Full replacement dashboard body (same shape as create).metadata (object, optional): Replacement metadata. Readonly system dashboards return a 403 error.Prefer this over update_dashboard for small refinements: unchanged panels are never re-sent. Reads the dashboard, applies ops in order and writes it back, so an edit made elsewhere between the read and the write is overwritten. Returns {dashboard_id, changes[]}; ops the dashboard already satisfies are left out, and if none changes anything nothing is written. The diff is checked against a re-read after the write: panel fields show stored values, not_stored lists keys the API dropped and not_removed keys it kept despite a null. Diff values over 200 characters are shortened in place, keeping their JSON type.
id (string, required): Dashboard UUID to patch.ops (array, required): Ops applied in order; if any is invalid nothing is written and the error names it as ops[i].
add_panel: panel (full panel object; version defaults to 1), optional index (0-based insert position, default append).update_panel: panel_id, fields (top-level panel keys to replace; null removes a key; id cannot change).remove_panel: panel_id.move_panel: panel_id, index (0-based position after the move).set_title: title (the dashboard name).set_variable: variable; replaces the variable with the same target, otherwise appends it.id (string, required): Dashboard UUID to delete. Readonly system dashboards cannot be deleted.dashboard_id (string, required): Dashboard UUID whose snapshots to list.Returns metadata only (id, name, expires_at, etc.). Use get_dashboard_snapshot for frozen panel data.
id (string, required): Snapshot UUID.Returns the full frozen snapshot including dashboard_definition, panel_data, time_range, and variables.
id (string, required): Snapshot UUID to delete.lookback_minutes / start_time_iso / end_time_iso (optional): Window; prefer lookback or explicit ISO bounds (default 60m).region (string, optional): Region override.Returns services that have profiling data in the window. Call this before get_flamegraph / get_top_functions / get_profile_summary.
service (string, required): Service name from get_profile_services.profile_type (string, optional): cpu (default), alloc, or wall. Pin a type when comparing windows.env / cluster / namespace / runtime (string, optional): Scope filters.limit (number, optional): Max aggregated stack rows (default 1000, max 10000).lookback_minutes / start_time_iso / end_time_iso / region (optional).Returns a nested flamegraph tree (name / value / self / children). truncated: true means the API row limit was hit.
Same filters as get_flamegraph. Returns self-sample ranking of hottest functions. May be truncated; check truncated.
Same filters as get_flamegraph. Returns a short natural-language triage of the profile for the service.
query (string, optional): Title substring. Empty lists broadly (subject to the 5,000-row cap).Returns {"dashboards":[…], "truncated":bool} with uid, title, uri, url, type, tags. Use uid with grafana_get_dashboard.
uid (string, required): Grafana dashboard uid.full_json (boolean, optional): When true, return raw Grafana JSON instead of the filtered summary.Default summary: version, tags, templating variables, and each panel's type/datasource/gridPos/promQL targets. Unknown plugin panel types appear in unsupportedPanelTypes.
No parameters. Returns the folder tree.
folder_uid (string, required): Grafana folder uid.Returns {"dashboards":[…], "truncated":bool} for dashboards in that folder (paginated up to 5,000).
No parameters. Returns a safe projection of datasources (no credential fields).
See TESTING.md for integration test setup and instructions.
Last9 MCP Server - Model Context Protocol server implementation for Last9
{ "mcpServers": { "last9-mcp-server": { "command": "npx", "args": [ "-y", "@last9/[email protected]" ] } }}Herramientas
Aún no hay lista de herramientas. Los servidores remotos la reciben tras la primera comprobación de estado.
Preguntas frecuentes
Copia en esta página la configuración stdio para Cursor, Claude Desktop, Claude Code o VS Code. Usa la instalación del publicador, no un proxy de PlugMCP.
Este registro no indica ninguna variable de entorno obligatoria.
@last9/ se ejecuta localmente por stdio.
Última comprobación: 2026-11-14. Last9 MCP Server - Model Context Protocol server implementation for Last9