Skip to main content
Glama
Azure DevOps MCP Server

Azure DevOps MCP Server

Azure DevOps MCP Server

🇮🇷 نسخه فارسی

A Model Context Protocol (MCP) server for Azure DevOps Server 2022 (on-premise) that exposes Azure DevOps capabilities as AI-callable tools — letting any MCP-compatible AI assistant query pipelines, pull requests, builds, repositories, and work items through natural language.

TypeScript Node.js MCP Docker Kubernetes


What can it do?

Ask your AI assistant in plain language — the MCP server handles the Azure DevOps REST API calls:

Simple queries

"List all projects in the collection"
"What branches does the backend repository have?"
"Show me all open pull requests"
"What's the status of the last build?"
"Who is assigned to work item #42?"
"Read the appsettings.json from the main branch"

Moderate queries

"List all failed builds in the last 24 hours and which pipelines they belong to"
"Show me the open PRs targeting the main branch and their reviewers"
"Find all active bugs in the project"
"Create a pull request from feature/payment to develop with a description"
"Add John as a reviewer to PR #87"
"Run the pipeline named 'deploy-staging' on the release branch"
"What was the last pipeline that ran on agent pool win19-prod-bi?"
"How many builds completed this week compared to last week?"
"How many build agents are online right now?"

Complex queries

"Which PRs opened in the last 48 hours still have no reviewer? Group them by repository."
"Get the logs of the last failed build for the 'deploy-prod' pipeline and summarize what went wrong"
"Find all In Progress work items assigned to me and list them by priority"
"Show me all unresolved review comments on PR #112 — what feedback is still pending?"
"How many completed builds ran in the last 7 days vs the 7 days before that? What is the percentage change?"
"Find high-priority work items that haven't been updated in more than 3 days"
"Which repositories have open PRs with no reviewer and at least one unresolved comment?"
"Which agent pools are available, and how many agents are online in each pool?"

Related MCP server: Azure DevOps MCP Server

Available Tools

Projects

Tool

Description

list_projects

List all projects in the collection

Repositories & Branches

Tool

Description

list_repositories

List Git repositories in a project

list_branches

List branches of a repository

get_file_content

Read file content from a repository at any branch

Pull Requests

Tool

Description

list_pull_requests

List PRs across all repos in a project (filter by status)

list_prs_without_reviewer

Find open PRs with no reviewer, with optional time window (e.g. last 24 hours)

get_pull_request

Get PR details: description, merge status, reviewers and their votes

get_pr_comments

Get review comment threads on a PR (filter by active/resolved)

add_pr_reviewer

Add a reviewer to a PR by email or display name

create_pull_request

Create a new PR from source to target branch (supports draft)

Pipelines & Builds

Tool

Description

list_pipelines

List all pipeline definitions in a project

list_agent_pools

List available agent pools/queues in a project

list_agents

List agents across all pools or a specific pool with online/offline status and counts

get_last_build

Get the most recent build — filter by pipeline name or agent pool name

list_builds

List builds with filters: pipeline (by ID or definitionNameFilter partial name match, e.g. "prod"), statusFilter (running state) and resultFilter (succeeded/failed/canceled), agent pool, and date range (minTime/maxTime). Includes webUrl per build and a truncated flag warning when older builds near minTime may be cut off by the top limit

list_failed_builds

Find failed/partial builds in the last N hours

get_build

Get details of a specific build

get_build_logs

Fetch console log output of a build (auto-truncated, last 150 lines). Set errorsOnly=true to return only error/warning lines — much smaller and faster for diagnosing failures

run_pipeline_by_name

Find and run a pipeline by name (partial match, no ID needed)

run_pipeline

Queue a pipeline run by numeric ID

Work Items

Tool

Description

list_work_items

Search work items by type, state, assignee, or keyword using WIQL

get_work_item

Get full details of a work item by ID

create_work_item

Create a new work item (Bug, Task, User Story, etc.)

update_work_item

Update fields of an existing work item (state, assignee, title, etc.)

Classic Release Pipelines

Tool

Description

list_release_definitions

List classic release pipeline definitions with their stages

list_releases

List releases with per-stage deployment status — filter by pipeline name and/or stage name (e.g. "last release deployed to Shatel")

get_release

Get full details of a release: all stage statuses, deploy times, and artifact versions

get_release_changes

Get the list of commits / TFVC changesets included in a release (answers "what changed in this deployment?")


Architecture

AI Client (Claude, Open WebUI, etc.)
        │  MCP Protocol (JSON-RPC)
        ▼
┌─────────────────────────┐
│   Azure DevOps MCP      │
│   ─────────────────     │
│  HTTP (Kubernetes) or   │
│  stdio (local)          │
│                         │
│  tools/                 │
│    projects.ts          │
│    repos.ts             │
│    pipelines.ts         │
│    workitems.ts         │
└────────────┬────────────┘
             │  REST API (api-version 7.0)
             │  Basic Auth (PAT)
             ▼
┌─────────────────────────┐
│  Azure DevOps Server    │
│  2022 (on-premise)      │
└─────────────────────────┘
  • Transport: Stateless Streamable HTTP for Kubernetes (scales horizontally, no sticky sessions) or stdio for local use

  • Auth: Personal Access Token via HTTP Basic auth (Authorization: Basic base64(:<PAT>))

  • API version: 7.0 — the highest supported by Azure DevOps Server 2022.0.x


Requirements

  • Node.js 20+

  • Azure DevOps Server 2022 (on-premise)

  • A Personal Access Token with:

    • Code (Read & Write) — for repo, branch, file, and pull request tools

    • Build (Read & Execute) — for pipeline and build tools

    • Work Items (Read & Write) — for work item tools


Quick Start

Local (stdio — for Claude Desktop / Claude Code)

npm install
npm run build

# Register with Claude Code
claude mcp add azure-devops \
  --env AZDO_ORG_URL=https://your-server.example.com/YourCollection \
  --env AZDO_PAT=your_pat_here \
  --env AZDO_PROJECT=YourProject \
  -- node dist/index.js

Docker

# Build
docker build -t your-registry/azure-devops-mcp:1.0.0 .

# Run locally for testing
docker run -p 3000:3000 \
  -e AZDO_ORG_URL=https://your-server.example.com/YourCollection \
  -e AZDO_PAT=your_pat_here \
  -e AZDO_PROJECT=YourProject \
  your-registry/azure-devops-mcp:1.0.0

Kubernetes

# 1. Create namespace
kubectl create namespace mcp-servers

# 2. Create secret (keep PAT out of git)
kubectl create secret generic azure-devops-mcp-secret \
  --namespace mcp-servers \
  --from-literal=AZDO_PAT='your_pat_here'

# 3. Edit k8s/configmap.yaml with your AZDO_ORG_URL and AZDO_PROJECT

# 4. Apply manifests
kubectl apply -f k8s/configmap.yaml
kubectl apply -f k8s/deployment.yaml
kubectl apply -f k8s/service.yaml

# 5. Verify
kubectl -n mcp-servers rollout status deploy/azure-devops-mcp

In-cluster endpoint:

http://azure-devops-mcp.mcp-servers.svc.cluster.local/mcp

Quick test with port-forward

# Terminal 1
kubectl -n mcp-servers port-forward deploy/azure-devops-mcp 3000:3000

# Terminal 2 — health check
curl http://127.0.0.1:3000/healthz

# Terminal 2 — list projects
curl -sS -X POST http://127.0.0.1:3000/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  --data-binary '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_projects","arguments":{}}}'

Configuration

Environment Variables

Variable

Required

Description

AZDO_ORG_URL

Yes

Collection URL, e.g. https://your-server.example.com/YourCollection

AZDO_PAT

Yes

Personal Access Token

AZDO_PROJECT

No

Default project — avoids passing project name in every request

AZDO_API_VERSION

No

Default: 7.0

MCP_TRANSPORT

No

http for Kubernetes, stdio for local (default: stdio)

PORT

No

HTTP port, default: 3000

MCP_PATH

No

Endpoint path, default: /mcp

.env example

AZDO_ORG_URL=https://your-server.example.com/YourCollection
AZDO_PAT=your_pat_here
AZDO_PROJECT=YourProject
MCP_TRANSPORT=http
PORT=3000

Project Structure

src/
├── index.ts          Entry point — selects transport based on MCP_TRANSPORT
├── config.ts         Env var reading and validation
├── server.ts         MCP server construction and tool registration
├── azureClient.ts    REST client with PAT Basic auth and 30s timeout
├── httpServer.ts     Stateless Streamable HTTP transport for Kubernetes
└── tools/
    ├── projects.ts   Project discovery tools
    ├── repos.ts      Repository, branch, and pull request tools
    ├── pipelines.ts  Pipeline and build tools
    ├── workitems.ts  Work item tools (WIQL, create, update)
    └── helpers.ts    Shared utilities (response formatting, truncation)

k8s/
├── configmap.yaml    Non-secret configuration
├── deployment.yaml   Kubernetes Deployment (non-root, read-only FS)
└── service.yaml      ClusterIP Service

Technical Notes

  • Stateless HTTP: Each request creates an independent MCP server instance — scales horizontally without sticky sessions.

  • Health check: GET /healthz for Kubernetes liveness and readiness probes.

  • API version 7.0: Azure DevOps Server 2022.0.x supports up to 7.0 only. Version 7.1 is available in Azure DevOps Services (cloud) and Server 2022.1+.

  • Response truncation: Tool responses are capped at 24,000 characters to avoid flooding the model context window.

  • Request timeout: All Azure DevOps API calls abort after 30 seconds with a clear error message.

  • Container security: Runs as non-root user 1000, read-only root filesystem, all Linux capabilities dropped.

  • Weak model friendly: Tools return pre-processed, ready-to-answer data rather than raw JSON — works well with smaller models that struggle to process large API responses.


License

MIT

Available Tools

16 tools
create_pull_requestCreate pull requestB

Create a new pull request from a source branch into a target branch.

Inputs: repositoryId (required), sourceBranch (required), targetBranch (required), title (required), +3 more.

get_buildGet buildB

Get details of a single build by id.

Inputs: buildId (required), project.

get_build_logsGet build logsA

Fetch the console log output of a build. Use this when the user asks for build logs, error details, failure reason, or what went wrong in a build. Returns log text per stage/task. Large logs are automatically truncated to the last 150 lines per entry.

Inputs: buildId (required), project, maxLinesPerEntry.

get_file_contentGet file contentB

Read the content of a file from a repository at an optional branch.

Inputs: repositoryId (required), path (required), project, branch.

get_last_buildGet last build statusA

Get the status and result of the most recent build in a project. Use this to answer questions like 'what is the last pipeline status?', 'did the last build succeed?', 'what ran last?'. Optionally filter by pipeline name (partial match). Returns: buildNumber, status (inProgress/completed), result (succeeded/failed/partiallySucceeded/canceled), pipeline name, branch, who triggered it, and start/finish time.

Inputs: project, pipelineName.

get_pull_requestGet pull requestA

Get details of a single pull request by id.

Inputs: repositoryId (required), pullRequestId (required), project.

list_branchesList branchesB

List branches (refs/heads) of a repository.

Inputs: repositoryId (required), project.

list_buildsList buildsB

List recent builds in a project, optionally filtered by pipeline/definition.

Inputs: project, definitionId, top, statusFilter.

list_failed_buildsList failed buildsA

Find builds that failed or partially succeeded in the last N hours. Use this for questions like 'آیا build شکست‌خورده‌ای داشتیم؟', 'failed builds in the last 24 hours', 'what broke today?', 'build failures this week'. Returns: pipeline name, result, branch, who triggered it, and when it ran.

Inputs: project, hours, includePartial, top.

list_pipelinesList pipelinesC

List pipelines defined in a project.

Inputs: project.

list_projectsList projectsA

List all projects in the Azure DevOps collection. Call this first when the project name is unknown or a lookup by name fails.

Inputs: no named parameters.

list_prs_without_reviewerList PRs without reviewerA

Find open pull requests that have no reviewer assigned, optionally within a time window. Use this for: 'which PRs have no reviewer?', 'unreviewed PRs in last 24 hours', 'PRs waiting for review this week'. Use the hours parameter to filter by creation date (e.g. hours=24 for last 24 hours). Omit hours to return all open PRs without a reviewer regardless of age.

Inputs: project, hours, top.

list_pull_requestsList pull requestsA

List pull requests across ALL repositories in a project, or filter to one repository. Use this for questions like 'show all open PRs', 'which PRs have no reviewer', 'list PRs in project X'. Leave repositoryId empty to search the whole project at once.

Inputs: project, repositoryId, status, top.

list_repositoriesList repositoriesB

List all Git repositories in a project.

Inputs: project.

run_pipelineRun pipelineA

Queue a new run of a pipeline by its numeric id. Use run_pipeline_by_name if you only know the name.

Inputs: pipelineId (required), project, branch.

run_pipeline_by_nameRun pipeline by nameA

Find a pipeline by name and run it. Use this when the user says 'run pipeline X', 'اجرا کن pipeline X', 'trigger X'. Searches for the pipeline by partial name match, then queues a run automatically. No need to call list_pipelines first.

Inputs: pipelineName (required), project, branch.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 16 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: PR creation vs retrieval, build logs vs build listing, specialized tools for failed builds or unreviewed PRs. There is no functional overlap that would cause ambiguity.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern in snake_case (e.g., create_pull_request, get_last_build, list_repositories). The naming is predictable and uniform across the entire set.

    Tool Count5/5

    16 tools cover core Azure DevOps operations (projects, repos, branches, PRs, builds, pipelines) without being bloated. Each tool serves a necessary function, and the count is appropriate for the domain.

    Completeness4/5

    The tool surface covers essential CRUD-like operations for PRs, builds, and pipelines. Minor gaps exist (e.g., no update or merge for PRs, no build cancellation), but the most common agent scenarios are supported.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers