Power your agent
คำสั่งของผู้เผยแพร่เอง ไม่ใช่เอนด์พอยต์ของ PlugMCP
Host a copy with our keys and logs.
npx -y @octopusdeploy/[email protected]Claude Code ไม่มีลิงก์ติดตั้งสำหรับคำสั่งแบบโลคัล ให้คัดลอกคำสั่งด้านล่างแล้วรันในเทอร์มินัล
claude mcp add octopusdeploy-mcp-server-github -- npx -y @octopusdeploy/mcp-server@2.3.3รันในเทอร์มินัล
{ "mcpServers": { "octopusdeploy-mcp-server-github": { "command": "npx", "args": [ "-y", "@octopusdeploy/[email protected]" ] } }}claude_desktop_config.json
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน 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.
เปิดจากผู้เผยแพร่
[!WARNING] Deprecated: This MCP server is deprecated. The Remote MCP Server is available in Octopus Server 2026.3 and later. See the Remote MCP Server documentation.
Octopus makes it easy to deliver software to Kubernetes, multi-cloud, on-prem infrastructure, and anywhere else. Automate the release, deployment, and operations of your software and AI workloads with a tool that can handle CD at scale in ways no other tool can.
Octopus Deploy Official MCP Server
Model Context Protocol (MCP) allows the AI assistants you use in your day to day work, like Claude Code, or ChatGPT, to connect to the systems and services you own in a standardized fashion, allowing them to pull information from those systems and services to answer questions and perform tasks.
The Octopus MCP Server provides your AI assistant with powerful tools that allow it to inspect, query, and diagnose problems within your Octopus instance, transforming it into your ultimate DevOps wingmate. For a list of supported use-cases and sample prompts, see our documentation.
Most tools exposed by the MCP Server use stable APIs that have been available from at least version 2021.1 of Octopus Server. Tools that are newer will specify the minimum supported version in the documentation. Alternatively, you can use the command line argument --list-tools-by-version to check how specific tools relate to versions of Octopus.
Credentials must be supplied via environment variables to avoid exposing them in the host process list (ps aux / /proc/<pid>/cmdline). The Octopus server URL can still be supplied via the --server-url flag.
docker run -i --rm -e OCTOPUS_API_KEY=your-key -e OCTOPUS_SERVER_URL=https://your-octopus.com octopusdeploy/mcp-server
Full example configuration (for Claude Desktop, Claude Code, and Cursor):
{
"mcpServers": {
"octopus-deploy": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"OCTOPUS_SERVER_URL",
"-e",
"OCTOPUS_API_KEY",
"octopusdeploy/mcp-server"
],
"env": {
"OCTOPUS_SERVER_URL": "https://your-octopus.com",
"OCTOPUS_API_KEY": "YOUR_API_KEY"
}
},
}
}
For Apple Mac users, you might need to add the following arguments in the configuration to force Docker to use the Linux platform:
"--platform",
"linux/amd64",
We are planning to release a native ARM build shortly so that those arguments will not be required anymore.
Full example configuration (for Claude Desktop, Claude Code, and Cursor):
Write tools enabled (default):
{
"mcpServers": {
"octopusdeploy": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@octopusdeploy/mcp-server"],
"env": {
"OCTOPUS_SERVER_URL": "https://your-octopus.com",
"OCTOPUS_API_KEY": "YOUR_API_KEY"
}
}
}
}
Read-only mode (recommended for production):
{
"mcpServers": {
"octopusdeploy": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@octopusdeploy/mcp-server", "--read-only"],
"env": {
"OCTOPUS_SERVER_URL": "https://your-octopus.com",
"OCTOPUS_API_KEY": "YOUR_API_KEY"
}
}
}
}
The Octopus MCP Server is typically configured within your AI Client of choice.
It is packaged as an npm package and executed via Node's npx command. Credentials (API key or access token) must be supplied via environment variables — they are not accepted as command-line arguments to avoid exposing secrets in the process list. The Octopus server URL may be supplied via either the OCTOPUS_SERVER_URL environment variable or the --server-url flag.
OCTOPUS_API_KEY=API-KEY \
OCTOPUS_SERVER_URL=https://your-octopus.com \
npx -y @octopusdeploy/mcp-server
Or with the server URL on the command line:
OCTOPUS_API_KEY=API-KEY \
npx -y @octopusdeploy/mcp-server --server-url https://your-octopus.com
The MCP server supports two authentication methods. Both are supplied via environment variables — credentials are not accepted on the command line because flags are visible in the host process list to any local user.
API keys are the standard authentication method for Octopus Deploy. You can generate one from your Octopus Deploy user profile.
OCTOPUS_API_KEY=API-XXXXXXXXXXXXXXXXXXXXXXXXXX \
OCTOPUS_SERVER_URL=https://your-octopus.com \
npx -y @octopusdeploy/mcp-server
The server also supports short-lived access tokens (Bearer tokens) as an alternative to API keys. This authentication method is intended only for automated scenarios where an external system issues a short-lived token to the MCP server (e.g., CI/CD pipelines, automated orchestration, or machine-to-machine workflows). Do not use long-lived Bearer tokens — use API keys instead for interactive or long-running sessions.
OCTOPUS_ACCESS_TOKEN=your-short-lived-token \
OCTOPUS_SERVER_URL=https://your-octopus.com \
npx -y @octopusdeploy/mcp-server
Full example configuration with an access token:
{
"mcpServers": {
"octopusdeploy": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@octopusdeploy/mcp-server"],
"env": {
"OCTOPUS_SERVER_URL": "https://your-octopus.com",
"OCTOPUS_ACCESS_TOKEN": "YOUR_TOKEN"
}
}
}
}
If both an API key and an access token are provided, the access token takes precedence. The active authentication method is recorded in the log file (configurable with --log-file) so operators can confirm which credential is in use.
The Octopus MCP Server supports several command-line options to customize which tools are available.
If you are not sure which tools you require, we recommend running without any additional command-line options and using the provided defaults.
Use the --toolsets parameter to enable specific groups of tools:
# Enable all toolsets (default)
npx -y @octopusdeploy/mcp-server
# Enable only specific toolsets
npx -y @octopusdeploy/mcp-server --toolsets projects,deployments
# Enable all toolsets explicitly
npx -y @octopusdeploy/mcp-server --toolsets all
Available toolsets:
The server runs with write tools enabled by default. Pass --read-only to disable all write tools and block POST/PUT/PATCH/DELETE through the execute backstop. Most curated tools are already read-only; only a small set perform writes.
Write-enabled tools (always-write):
create_release - Create new releasesdeploy_release - Deploy releases to environments and tenantsrun_runbook - Run a runbook against one or more environments (and optional tenants)update_feature_toggle - Adjust per-environment state and rollout percentages on an existing feature toggleConditionally-writing tool: execute is a structured REST backstop whose tier (read / write / delete) is determined by the HTTP method passed to it. See the API Catalog & Backstop section for details.
Write tools are gated by an MCP elicitation prompt: clients that support elicitation will be asked to confirm before the call proceeds. Clients without elicitation support must pass confirm: true in the tool arguments — otherwise the tool aborts with an error. Set OCTOPUS_SKIP_ELICITATION=true to bypass the gate entirely (intended for unattended automation).
The server uses a three-tier read/write/delete classification, enforced server-side based on the HTTP method (the agent cannot bypass this by lying about intent):
execute, plus all find_* / get_* / list_* tools.execute and the always-write tools above. Blocked when --read-only is set.execute. Requires --allow-deletes and is blocked when --read-only is set. A small set of catastrophic-delete paths (e.g. DELETE /api/spaces/{id}, DELETE /api/users/{id}) and API-key endpoints are on a hard sensitive denylist that ignores both flags.# Default - write tools enabled (POST/PUT/PATCH)
npx -y @octopusdeploy/mcp-server
# Additionally permit DELETE requests through the execute tool
npx -y @octopusdeploy/mcp-server --allow-deletes
# Read-only mode - write/delete tools disabled
npx -y @octopusdeploy/mcp-server --read-only
Security Note: Use an API key with appropriate, least-privilege permissions — write operations can create releases and trigger deployments in your Octopus instance. For production, consider passing --read-only unless you have a specific, controlled use case for writes. --allow-deletes is off by default; only enable it when the agent must issue DELETE requests through execute. If you pass --allow-deletes together with --read-only, the server prints a startup warning to stderr — DELETE requests remain blocked by the read-only gate.
All examples below assume OCTOPUS_API_KEY is set in the environment. The --server-url flag is shown for clarity but can also be provided via OCTOPUS_SERVER_URL.
# Development setup with only core and project tools
npx -y @octopusdeploy/mcp-server --toolsets core,projects --server-url https://your-octopus.com
# Production setup with all tools and read-only enforcement
npx -y @octopusdeploy/mcp-server --toolsets all --read-only --server-url https://your-octopus.com
# Default invocation - all tools and writes enabled
npx -y @octopusdeploy/mcp-server --server-url https://your-octopus.com
--read-only - Enable read-only mode: disable all curated write tools and block POST/PUT/PATCH/DELETE through execute. Writes are enabled by default; this flag turns them off. See Read-Only Mode.--allow-deletes - Permit DELETE requests through the execute tool. Ignored (with a startup warning) when --read-only is set. Default false.--log-level <level> - Minimum log level (info, error)--log-file <path> - Log file path or filename. If not specified, logs are written to console only-q, --quiet - Disable file logging, only log errors to console--list-tools-by-version - List all registered tools by their supported Octopus Server version and exitQuick start: Paste Octopus URLs directly to investigate issues without manual ID extraction.
get_deployment_from_url: Get deployment details from deployment URL (returns taskId for follow-up)get_task_from_url: Get task details and logs from task URLDeployment investigation workflow:
1. get_deployment_from_url with deployment URL
→ Returns deployment context + taskResourceUri + grepTaskLogHint
2a. Fetch the structured activity tree via resources/read (or read_resource)
octopus://spaces/{spaceName}/tasks/{taskId}/details
2b. Or call grep_task_log with the taskId to search the raw log without
fetching the full body:
grep_task_log({ spaceName, taskId, pattern: "error|fail", caseInsensitive: true })
Task investigation (direct task URL):
get_task_from_url with task URL
→ Returns task details and logs immediately
These tools eliminate manual ID extraction by:
Example URLs:
https://your-octopus.com/app#/Spaces-1/projects/my-app/deployments/Deployments-123https://your-octopus.com/app#/Spaces-1/tasks/ServerTasks-456See Working with URLs for detailed workflows, examples, and best practices.
list_spaces: List all spaces in the Octopus Deploy instancelist_environments: List all environments in a given spaceThese tools and resources let the agent reach Octopus REST endpoints that don't have a dedicated curated tool, with hard server-side gating between read, write, and delete operations.
grep_llms_txt: Search the Octopus API catalog (octopus://api/llms.txt) with grep-style semantics (minimum supported Octopus version: 2026.2.3916). The catalog body is large (typically 300+ KB) — call this rather than reading the resource body directly. Parameters mirror GNU grep (pattern, caseInsensitive, invertMatch, fixedString, beforeContext, afterContext, maxCount). Useful for discovering endpoints (POST /releases), enumerating delete endpoints (DELETE ), or finding the body type for a write operation (Body: Create.*Command).execute: Structured REST backstop. Reaches any Octopus REST endpoint under /api. The HTTP method is the authoritative read/write/delete classifier — never an isWrite flag the LLM can set. Method gating is hard-coded server-side:
GET is always allowed (subject to the path shape check + sensitive denylist).POST/PUT/PATCH are blocked when --read-only is set; otherwise they require user confirmation via elicitation.DELETE requires --allow-deletes (and is blocked when --read-only is set) plus a stronger "IRREVERSIBLE" elicitation message.DELETE /api/spaces/{id}, DELETE /api/users/{id}) is enforced even with both flags on./api or start with /api/ — absolute URLs, SDK-relative ~/api/... paths, and host-relative paths outside /api (e.g. /octopus/portal/...) are rejected up front, so execute stays bounded to the Octopus REST API surface.--toolsets has been narrowed. With every toolset enabled (the default, or explicit --toolsets all) the allowlist is bypassed and any path under /api is reachable subject to the gates above. When --toolsets is narrowed the allowlist becomes the kill-switch: paths only resolve if their owning toolset is enabled, so disabling a toolset (e.g. certificates) makes its paths unreachable through execute even on GET.Catalog data is also exposed as MCP Resources:
octopus://api/llms.txt — markdown catalog of every Octopus REST endpoint (HTTP method, path, query params, request/response types). Requires Octopus Server 2026.2.3916 or later. 5-minute in-memory cache keyed on the configured server URL. Prefer grep_llms_txt to reading the body directly.octopus://api/capabilities — JSON describing the running session: server version, enabled toolsets, available tools (with their minimumOctopusVersion), and whether --read-only / --allow-deletes is on. Useful for the agent to discover what's reachable in this session.list_projects: List all projects in a given spacedeploy_release: Deploy a release to environments (supports both tenanted and untenanted deployments)list_deployments: List deployments in a space with optional filteringcreate_release: Create a new release for a projectfind_releases: Find releases in a space (can get a specific release by ID, or list/filter releases by project)Release detail is also available as an MCP Resource at octopus://spaces/{spaceName}/releases/{releaseId} — fetch via resources/read (or the read_resource backstop tool) to get the full release body, including release notes and selected packages.
find_runbooks: Find runbooks in a project (can get a specific runbook by ID, or list/filter runbooks by partial name). Each summary includes the published snapshot ID, multi-tenancy mode, and environment scope so callers can pick valid targets before running.run_runbook: Run a runbook against one or more environments. Supports tenanted runs (by tenant name or tenant tag), prompted variables, guided failure mode, scheduled run windows, and step or machine inclusion/exclusion. Defaults to the runbook's published snapshot if runbookSnapshotId is omitted.The full runbook body (including runtime policy fields) is available as an MCP Resource at octopus://spaces/{spaceName}/runbooks/{runbookId}.
Task data is primarily exposed as MCP Resources. Use resources/read (or the read_resource backstop tool) with one of:
octopus://spaces/{spaceName}/tasks/{taskId} — lightweight metadata (state, timing, completion flags)octopus://spaces/{spaceName}/tasks/{taskId}/details — full ServerTaskDetails (Progress, ActivityLogs tree, etc.)For log search, use the grep_task_log tool rather than a /log resource:
grep_task_log: Search a task's activity log without fetching the full body. Parameters mirror GNU grep (pattern, caseInsensitive, invertMatch, fixedString, beforeContext, afterContext, maxCount). Returns matching lines with 1-indexed lineNumber, optional before/after context arrays, and a totalMatches count across the whole log.There is intentionally no /log resource: activity logs can be multi-megabyte, and an addressable resource would tempt callers to fetch the entire body when grep is almost always the right primitive.
find_tenants: Find tenants in a space (can get a specific tenant by ID or list/search tenants with filters)get_tenant_variables: Get tenant variables by type (all, common, or project)get_missing_tenant_variables: Get tenant variables that are missing valuesget_kubernetes_live_status: Get live status of Kubernetes resources for a project and environment (minimum supported version: 2025.3)find_deployment_targets: Find deployment targets in a space (can get a specific target by ID or list/search targets with filters)find_certificates: Find certificates in a space (can get a specific certificate by ID or list/search certificates with filters)find_accounts: Find accounts in a space (can get a specific account by ID or list/search accounts with filters)find_interruptions: Find pending or historical interruptions (manual interventions, approvals, guided-failure prompts) in a space, optionally filtered by task, project, environment, regarding document, responsibility, or pending state. Returns slim summaries; dereference the octopus://spaces/{spaceName}/interruptions/{interruptionId} resource for the full Form definition (control types, Markdown instructions, button options, submitted Form.Values).find_feature_toggles: List customer feature toggles in a project. Each summary includes per-environment state (isEnabled, rolloutPercentage, clientRolloutPercentage) plus a resourceUri so "where is X turned on" is answerable from the list response.update_feature_toggle: Adjust an existing toggle. Narrow surface — flip an environment on/off, change rollout percentages, or update the toggle-level description / default state. Internally fetches the current toggle, applies your patches in memory, and PUTs the merged body, so unmentioned environments and unmentioned fields are preserved. Patches that reference an environment not already configured on the toggle are rejected.The full toggle body (description, tenants, segments, minimum versions) is available as an MCP Resource at octopus://spaces/{spaceName}/projects/{projectId}/featuretoggles/{slug}. Rollout group bodies are addressable at octopus://spaces/{spaceName}/projects/{projectId}/rolloutgroups/{rolloutGroupId} for read-only inspection.
Out of scope (use the Octopus UI): creating new feature toggles, deleting toggles, renaming or retagging, attaching/detaching rollout groups, tenant targeting, segments, minimum-version filters, and rollout-group / SDK client-identifier management.
get_deployment_process: Get deployment process by ID for projects or releasesget_variables: Get all project variables and library variable set variables for a project (supports config-as-code projects via gitRef)get_branches: Get Git branches for a version-controlled project (minimum supported version: 2021.2)get_current_user: Get information about the current authenticated userThe Octopus MCP Server includes both read and write operations. Important security considerations:
By default, the following write operations are available:
execute backstop: Bounded to paths under /api, with an always-on sensitive denylist. The per-toolset path allowlist applies only when --toolsets has been narrowed; with all toolsets enabled (the default) the only path gates are the /api boundary and the sensitive denylist.Pass --read-only to disable all of the above. DELETE requests through execute require an additional --allow-deletes flag — a deliberate opt-in for irreversible operations — and remain blocked when --read-only is set.
Critical Security Measures:
--read-only unless you have a specific, controlled use case for write operations. DELETE always requires the additional --allow-deletes opt-in.execute is the authoritative classifier. The agent cannot bypass the gate by misrepresenting what the call does — POST/PUT/PATCH/DELETE requests get tier-specific gating regardless of the prose in the request body.--toolsets removes both the disabled toolsets' curated tools and their paths from the execute allowlist. (The allowlist is only consulted when toolsets are narrowed; with all toolsets enabled execute is bounded by the /api shape check and the sensitive denylist instead.)Recommendation: For production environments, pass --read-only unless you have a specific, controlled use case for write operations. Leave --allow-deletes off unless you specifically need DELETE semantics through execute.
The nature of current AI chat tools and the MCP protocol itself makes it impractical to analyze large amounts of data. Most MCP clients currently do not support chaining tool calls (using the output of one tool as input to the next one) and instead fall back to copying the results token by token, which frequently leads to hallucinations. If you are looking to process historical data from your Octopus instance for analysis purposes, we recommend using the API directly or writing your own MCP client that is capable of processing the tool call results programmatically.
The MCP Server is technically just a thin layer on top of the existing Octopus Server API. As such it is capable of retrieving large amounts of data (for example, requesting thousands of deployments). Such queries can have a significant effect on your instance's performance. Instruct your models to only retrieve the minimum set of data that it needs (most models are really good at this out of the box).
Contributions are welcome! :heart: Please read our Contributing Guide for information about how to get involved in this project.
We are eager to hear how you plan to use Octopus MCP Server and what features you would like to see included in future version.
Please use Issues to provide feedback, or request features.
If you are a current Octopus customer, please report any issues you experience using our MCP server to our support team. This will ensure you get a timely response within our standard support guarantees.
We are working on integrating an MCP server directly into Octopus Server. This will open up the door for us to build more complex MCP tools, as well as:
If this is of interest to you, please register your interest on our roadmap item.
This project is licensed under the terms of Mozilla Public License 2.0 open source license.
Octopus Deploy Official MCP Server
{ "mcpServers": { "octopusdeploy-mcp-server-github": { "command": "npx", "args": [ "-y", "@octopusdeploy/[email protected]" ] } }}เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
OctopusDeploy ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-11-10 Octopus Deploy Official MCP Server