azure-devops-mcp

azure-devops-mcp

Multi-user wrapper around Microsoft's Azure DevOps MCP server, enabling AI agents to authenticate per-user and interact with Azure DevOps tools via isolated PATs.

Category
Visit Server

README

Azure DevOps MCP (multi-user wrapper)

A self-hosted, single-container service that exposes Microsoft's official Azure DevOps MCP server to AI agents over an authenticated remote endpoint. It handles auth (passkeys for people, OAuth 2.1 / API tokens for agents) and per-user Azure DevOps PATs, and passes MCP tool calls straight through to Microsoft's server.

It is multi-user with no admin: anyone who can reach it self-registers and manages their own connections. A user's PAT is never reachable by another user.

How it works

Microsoft's @azure-devops/mcp is a stdio process that holds exactly one PAT (PERSONAL_ACCESS_TOKEN, base64 email:pat) and targets one organization (a positional arg), via --authentication pat. A shared child therefore cannot serve multiple users without leaking PATs. So this wrapper:

  • spawns one Microsoft-MCP child per connection ({org, PAT}), lazily and kept warm, with that connection's PAT injected into only that child's env;
  • fronts agents with a single /mcp endpoint (Streamable HTTP + bearer auth) that, per request, builds a proxy bound to exactly one authenticated user;
  • forwards tools/* and prompts/* to the user's children, namespaced as <slug>__<tool>, and audits every call.
agent ──Bearer──▶ /mcp ──(user-scoped proxy)──▶ child(orgA, userX PAT)
                                              └─▶ child(orgB, userY PAT)

The never-bleed guarantee

  • Every bearer credential resolves to a user_id. The proxy is built per request and bound to that one user.
  • tools/list only iterates that user's connections; tools/call re-validates the slug belongs to the user before routing (src/mcp/proxy.ts).
  • Children are keyed by connection_id, which is only ever obtained via WHERE user_id = ?. There is no code path from one user's request to another user's connection or PAT.
  • PATs are encrypted at rest with AES-256-GCM under MASTER_KEY (src/crypto.ts), decrypted only in memory at child spawn, never logged.
  • test/isolation.integration.test.ts boots the app with two users and proves a user can't list or call another user's tools, even by guessing the slug.

Live status wall

The root path / is a public, no-auth dashboard showing aggregate, anonymous activity across all accounts: lifetime tool calls, calls/min, p50/p95 latency, error rate, a 60-second sparkline, active upstream children, and a live feed of recent calls (generic upstream tool names only — never users, orgs, slugs, args, or PATs). It updates over SSE (/status/stream); /status/json is the snapshot.

Counters live in memory (src/metrics.ts), flush to the DB every 30s (lifetime totals survive restarts), and push to connected browsers every 2s.

Quick start (dev)

bun install
mise run dev   # http://localhost:3000

mise run dev sets ALLOW_DEV_MASTER_KEY=1, which opts into a built-in "DEVELOPMENT" key so locally stored PATs survive restarts. Any deployment without MASTER_KEY and without that explicit flag refuses to boot — so a misconfigured server can never silently encrypt real PATs with the public dev key.

Open /register, create a passkey, then add a connection (org + PAT) on the Connections page. Create an API token on the Tokens page and point a client at it:

claude mcp add --transport http azure-devops http://localhost:3000/mcp \
  --header "Authorization: Bearer <token>"

Ask the agent to call list_connections, then <slug>__<tool>.

Deploy

Single container. Put it behind a TLS-terminating proxy/tunnel that sets X-Forwarded-Proto/X-Forwarded-Host (passkeys need a secure origin). Sample composes in deploy/:

Open registration: anyone who can reach the URL can create an account. PATs stay isolated per user, but the network boundary is your real perimeter — front it with a VPN / private ingress / Cloudflare Access.

Back up MASTER_KEY. Losing it makes every stored PAT unrecoverable.

Configuration

Env Default Purpose
MASTER_KEY (required) 32-byte base64/hex key; encrypts PATs at rest. Required unless ALLOW_DEV_MASTER_KEY=1
ALLOW_DEV_MASTER_KEY 0 Local dev only: opt into the built-in insecure key when MASTER_KEY is unset
PUBLIC_URL unset (derive from proxy) Hard-pin the public origin/rpID
PORT 3000 Listen port
ADO_MCP_VERSION latest Upstream @azure-devops/mcp version to spawn
ADO_MCP_DIST_TAG latest npm dist-tag the updater tracks
ADO_MCP_AUTO_UPDATE 1 Auto-adopt newer upstream versions and recycle children
UPDATE_CHECK_INTERVAL_MS 7200000 How often to poll npm for a newer upstream (2h)
CHILD_IDLE_MS 600000 Reap idle upstream children after this
AUTH_RESET 0 Wipe all passkeys + sessions on boot

Upstream auto-update

A background checker polls npm for @azure-devops/mcp every couple of hours and, by default, adopts any newer version automatically — recycling all children onto it. There's no user action: this is a multi-user service, so upgrades just happen (src/updater.ts). Pin a version with ADO_MCP_VERSION or set ADO_MCP_AUTO_UPDATE=0 to disable automatic adoption.

Development

mise run test        # bun test
mise run typecheck   # tsc --noEmit
mise run build       # CSS + bundle to dist/
mise run docker:build

Tests use test/fake-ado-mcp.ts — a tiny stdio MCP server that stands in for Microsoft's, so no real org/PAT is needed.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured