Power your agent
การติดตั้งจากผู้เผยแพร่ ไม่ใช่เอนด์พอยต์ของ PlugMCP
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.
เปิดOpening an Azure support ticket should take seconds, not minutes. azure-support-ticket-mcp is a fast, local MCP server that turns the full ticket lifecycle into a guided conversation — no portal, no context switch, no re-typing the resource ID Azure already knows about.

macOS / Linux:
curl -sSL https://github.com/artlovan/azure_support_ticket_mcp/releases/latest/download/install.sh | sh
Windows (PowerShell):
irm https://github.com/artlovan/azure_support_ticket_mcp/releases/latest/download/install.ps1 | iex
The install script detects your OS and CPU architecture, downloads the matching binary from the latest GitHub release, verifies its SHA256 checksum, and installs it under your home directory:
$HOME/.local/bin/azure-support-ticket-mcp
(override with --prefix=/usr/local for a system-wide install).%LOCALAPPDATA%\Programs\azure-support-ticket-mcp\.The script prints a one-line PATH hint if the install directory isn't
already on your PATH. Open a new shell after installing so the change
takes effect.
Prefer to download the binary yourself? Every release also publishes the raw per-platform binaries and their
.sha256sidecars athttps://github.com/artlovan/azure_support_ticket_mcp/releases/latest/download/azure-support-ticket-mcp-<os>-<arch>. Pick the asset matching your platform, verify the checksum,chmod +x, and move it onto yourPATH.
Verify it's installed and the embedded support-service catalog loaded correctly (no Azure call yet):
azure-support-ticket-mcp doctor
You should see something like:
app dir: /Users/you/.azure-support-ticket-mcp
cache path: /Users/you/.azure-support-ticket-mcp/cache.sqlite
cloud: AzurePublicCloud
drafts.store: memory
seed download: false
cache: OK (425 services, seed version "2024-04-01-1")
az cli: FOUND (/opt/homebrew/bin/az)
arm reachable: OK (HTTP 400 Bad Request)
cache: OK (425 services …) — the embedded support-service catalog
loaded; the server is usable offline.az cli: FOUND — if present, the server can fall back to the az CLI
for auth (so you don't have to set service-principal env vars).arm reachable: OK (HTTP 400 …) — 400 on a HEAD probe is expected and
proves we can talk to management.azure.com.Building from source instead? See
CONTRIBUTING.md→ Development environment.
Either:
# Easiest: use the Azure CLI you already have.
az login
…or set environment variables for a service principal:
export AZURE_TENANT_ID=...
export AZURE_CLIENT_ID=...
export AZURE_CLIENT_SECRET=...
The server prefers env vars and falls back to az unless you set
AZURE_SUPPORT_TICKET_MCP_AUTH_ALLOW_AZ_CLI_FALLBACK=false.
copilot plugin install artlovan/azure_support_ticket_mcp:plugin
This installs a small Copilot CLI plugin that tells Copilot
how to spawn the binary you installed in step 1 (azure-support-ticket-mcp serve).
The plugin itself is just a launcher config — it does not ship the
binary, so step 1 must be done first.
Verify with:
copilot plugin list # should show azure-support-ticket-mcp
copilot mcp list # should show azure-support-ticket-mcp registered
Prefer to register manually without the plugin? You can still do it the old-fashioned way — the plugin is purely a convenience wrapper:
copilot mcp add azure-support-ticket-mcp -- \ "$(command -v azure-support-ticket-mcp)" serve
copilot mcp addonly registers the launcher in Copilot's config — it does not start anything. The MCP process is spawned by Copilot CLI each time you open an interactive session and is shut down when the session ends. Inspect withcopilot mcp list· remove withcopilot mcp remove azure-support-ticket-mcp.
Tip: add
--allow-tool=azure-support-ticket-mcpto anycopilotcommand below to skip the per-tool approval prompts. The MCP's own preview-then-confirm gate still requires you to actively approve every state-changing call — see Security.
In a new Copilot CLI session, try a prompt like any of these:
# AKS cluster scale-out problem
copilot -i "open a support ticket — my AKS cluster prod-aks can't scale out"
# Azure OpenAI rate limiting
copilot -i "my Azure OpenAI deployment gpt-4o-prod is returning 429s every few minutes — please open a ticket"
# GPU VM quota error
copilot -i "file a support request — I can't deploy more GPU VMs in eastus, hitting a quota error"
# Networking / private endpoint issue
copilot -i "open a ticket — VMs in app-vnet can't reach Cosmos DB prod-cosmos through its private endpoint"
Or pipe an error log straight in from your terminal:
# Any log file — generic catch-all
copilot -i "ticket this: $(cat err.log)"
# GitHub Actions / CI failure
copilot -i "ticket this: $(gh run view <run-id> --log-failed)"
# Azure Developer CLI: provision + deploy
copilot -i "ticket this: $(azd up 2>&1)"
# Azure Functions deploy failure
copilot -i "ticket this: $(func azure functionapp publish my-app 2>&1)"
# Live app logs from Azure Container Apps
copilot -i "ticket this: $(az containerapp logs show -n my-app -g rg --tail 200 2>&1)"
# ARM deployment failure
copilot -i "ticket this: $(az deployment operation list --resource-group rg --name dep -o json 2>&1)"
# Pod scheduling / events / image-pull issues
copilot -i "ticket this: $(kubectl describe pod my-pod 2>&1)"
# Runtime app logs from a pod
copilot -i "ticket this: $(kubectl logs my-pod --tail=200 2>&1)"
# Terraform IaC apply failure
copilot -i "ticket this: $(terraform apply 2>&1)"
Copilot will walk you through: pick the right support service → pick a problem classification → fill contact details → preview → confirm → ticket opens. When you pipe error logs, the MCP auto-extracts the safe context (resource IDs, error codes, severity hints) and runs the LLM sanitization step described in Security before anything is persisted or sent to Azure.
Every side-effecting tool (
create_support_ticket,update_support_ticket,reply_to_ticket,add_attachments_to_ticket) is gated: nothing hits Azure until you confirm a one-timereview_token+draft_hashreturned by the preview call. Stale or tampered confirmations are rejected.
The intent is to make piping raw error output into a support ticket safer than copy-pasting it through the portal, not to make it risk-free. A few deliberate choices help here:
When you paste or pipe error text, the LLM in your host (Copilot CLI, Claude, etc.) reads it first and is instructed to scrub obvious secrets (connection strings, keys, tokens, private keys, passwords) before passing the result to the MCP. The MCP itself never receives the raw text directly as the draft body — it gets the sanitized version plus a summary of what was redacted.
This means the LLM does see your raw text on the way through. Treat the LLM as part of your trust boundary (which it already is for any other Copilot interaction); the MCP doesn't change that.
Before saving a sanitized draft, the MCP runs a small unambiguous-pattern check (e.g. PEM private key blocks, full Azure storage connection strings) and refuses to persist the draft if any matches. This is a safety net, not a guarantee — it covers a deliberately narrow set of patterns that should never appear in a support ticket regardless of context. Anything subtler (API keys without distinguishing prefixes, custom-format tokens, business data) relies on the LLM step above.
Every state-changing tool — create, update, reply, attach — is two-call. The first call returns a preview showing exactly what will hit Azure. The second call submits it, and only goes through if the draft hasn't changed in between.
~/.azure-support-ticket-mcp/): cache of tenant /
subscription / service / classification metadata and ticket history (no
secrets); your contact templates (name, email, phone, locale — whatever
you put in them).To wipe all local state (cache, templates, drafts), delete the app directory. The MCP will re-create what it needs on next startup; you'll lose your saved contact templates and need to re-seed them.
rm -rf ~/.azure-support-ticket-mcp/
copilot -i "ticket this: $(cat err.log)" walks all the way to a draftcopilot -p "..." --allow-all-tools (the safety contract still holds; the assistant just runs preview -> submit back-to-back)The full ticket lifecycle — open, manage, converse, attach, close — through one set of tools.
Lifecycle:
top / state filters; cached locally)Conversation with support engineers:
top=1) for "what did support say last?"Attachments (images, logs, anything):
default template every time you create a ticket (opt-out per call)az login credential chain — zero new credentials to managewhoami — shows the signed-in identity (email/UPN) the MCP will useazure_auth_status — explicit check of token validity / which credential type is activerefresh_support_cache for explicit warm-up after a long idledoctor — one-shot self-check (auth, cache, ARM reachability, file workspace API)See the Reusable contact templates feature group for the full capability list — this section is the setup walkthrough.
The MCP can remember your contact details (name, email, country, language,
timezone, etc.) so you're not asked the same questions every time you open a
ticket. Templates live in ~/.azure-support-ticket-mcp/templates/<name>.json.
default template (auto-created)The MCP creates a default template automatically the first time it runs.
It seeds the template from:
primary_email_address, first_name, last_name, tenant_id;LANG, /etc/localtime) → country, preferred_support_language, preferred_time_zone.Inspect what got seeded:
cat ~/.azure-support-ticket-mcp/templates/default.json
Update it through Copilot — "set my preferred contact method to phone and
add +1-555-0100" calls save_ticket_template and patches the JSON. Or
edit the file directly. To re-seed from scratch (e.g. after switching
identities), use "reset my default template" (init_ticket_template with
overwrite=true).
When you want a separate template — say, on-call contact for a specific service — save one:
"Save a template called
team-akswith on-call email[email protected]and description 'On-call for AKS prod'."
Copilot calls save_ticket_template. Then later:
"Open a critical ticket using the
team-akstemplate — pod crash loop onprod-aks-east."
Copilot passes template_name: "team-aks" to start_support_ticket_flow. Any
fields the template provides won't be re-asked.
After a successful create_support_ticket, the contact slice of the submitted
draft is written back to default.json (best-effort, non-fatal). To opt out
for one ticket: "don't update my default template this time"
(save_as_default_template: false). To capture under a named template:
"save these contacts as team-aks" (save_as_template_name: "team-aks").
Template names must match [A-Za-z0-9_-]{1,64} (blocks path traversal).
Writes are atomic (temp file + rename). For the full list of template tools see
Tool reference.
You can paste these requests into Copilot CLI — the model picks the right tools under the hood. The tool names below let you trace what's happening if you want to follow along in logs.
"Open a moderate severity ticket about my prod AKS cluster failing to autoscale. Use my default subscription."
Behind the scenes:
resolve_issue_context ranks support services from the resource id / phrase.list_problem_classifications lists problem buckets for the chosen service.start_support_ticket_flow → build_ticket_draft fills the form.preview_ticket_draft shows you the body — confirm before submission.create_support_ticket PUTs it; returns ticket name + portal URL +
a copy-paste-friendly share message."Same as above, but also attach
~/logs/aks-events.logand~/logs/kubectl-describe.txt."
Adds one extra step before submission:
prepare_attachments — creates the file workspace, uploads each file (chunked
to ≤2.5 MB base64 per chunk; files capped at 5 MB), and pins the workspace
name to the draft. create_support_ticket then reuses that name as the
ticket name (Azure convention)."What's the latest on ticket
xxxxxxxxxxxxxx-xxxx?"
summarize_ticket_thread — pulls ticket body + recent communications,
produces a deterministic local summary (counts inbound/outbound replies,
truncates the latest message body to ~500 chars). No LLM is invoked from
inside the MCP.For finer detail:
list_ticket_communications — paged thread (max 10 per page; pass top=1
to get just the newest reply).get_support_ticket — full raw body."Reply to ticket
xxxxxxxxxxxxxx-xxxx: subject 'Update', body 'Issue still reproducing in eastus2, attaching new logs'."
reply_to_ticket — first call returns a preview. Copilot shows you exactly
what will be posted; confirming sends it."Add
~/logs/new-trace.txtto ticketxxxxxxxxxxxxxx-xxxx."
add_attachments_to_ticket — same two-call confirmation pattern. Note that
attachments live on the ticket workspace, not on any individual reply."Bump ticket
xxxxxxxxxxxxxx-xxxxto critical and set primary email to[email protected]."
update_support_ticket — two-call. Allowed fields: severity, status,
advancedDiagnosticConsent, contact details."List my open support tickets in the prod subscription."
list_support_tickets — paged via $top + next_link, optional OData
$filter like Status eq 'Open'.All settings live in ~/.azure-support-ticket-mcp/:
Env overrides (prefix AZURE_SUPPORT_TICKET_MCP_):
Azure auth chain: env vars (AZURE_TENANT_ID + AZURE_CLIENT_ID +
AZURE_CLIENT_SECRET) → az CLI fallback (when enabled).
The MCP advertises all tools and their JSON schemas over the protocol itself — your client (Copilot CLI, Claude Desktop, etc.) lists them on connect. For engineering-level docs see docs/ARCHITECTURE.md; for the canonical tool descriptions read the #[tool] annotations in src/mcp/server.rs.
Contributions are welcome. See CONTRIBUTING.md for the
development environment setup, Rust style, MCP tool contract, testing
expectations, and PR workflow. For larger changes, please open an issue
first to discuss the approach.
For engineering guardrails see
.github/copilot-instructions.md;
for the full design see docs/ARCHITECTURE.md;
for shipped vs planned work see docs/ROADMAP.md.
MCP server that turns the full ticket lifecycle into a guided conversation — no portal, no context switch, no re-typing the resource ID Azure already knows about.
เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Azure Support Ticket ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-10-28 MCP server that turns the full ticket lifecycle into a guided conversation — no portal, no context switch, no re-typing the resource ID Azure already knows about.