Bitbucket MCP Server

MCP server for Bitbucket API operations. Works with Claude Code, Claude Desktop, Cursor, and any MCP-compatible client.

Language Versions

This repository contains both TypeScript and Python implementations:

Note: The TypeScript version is used for Smithery deployments. Both versions provide identical functionality.

Features

• Repositories: get, create, delete, list, update (move to project, rename)
• Pull Requests: create, get, list, merge, approve, decline, request changes, comments, diff
• Pipelines: trigger, get status, list, view logs, stop
• Branches: list, get
• Projects: list, get
• Commits: list, get details, compare/diff between branches
• Commit Statuses: get build statuses, create status (CI/CD integration)
• Deployments: list environments, get environment details, deployment history
• Webhooks: list, create, get, delete
• Tags: list, create, delete
• Branch Restrictions: list, create, delete branch protection rules
• Source Browsing: read files, list directories without cloning
• Repository Permissions: manage user and group permissions
• Pipeline Variables: manage CI/CD environment variables
• MCP Prompts: reusable workflow templates (code review, release notes, etc.)
• MCP Resources: browsable workspace data

Quick Start

TypeScript (Recommended for Smithery)

# Install globally
npm install -g mcp-server-bitbucket

# Or run directly with npx
npx mcp-server-bitbucket

Python

# Install with pipx
pipx install mcp-server-bitbucket

# Configure Claude Code
claude mcp add bitbucket -s user \
  -e BITBUCKET_WORKSPACE=your-workspace \
  -e BITBUCKET_EMAIL=your-email@example.com \
  -e BITBUCKET_API_TOKEN=your-api-token \
  -- mcp-server-bitbucket

Full Installation Guide - Includes API token creation, permissions setup, and troubleshooting.

Hosted

If you'd rather not run the server yourself, a hosted deployment is available on Fronteir AI.

Configuration

Environment Variables

Authentication modes

Bitbucket Cloud has two credential families, and they do not share an auth scheme:

Access tokens return 401 with Basic auth, so the server picks the mode automatically:

1. BITBUCKET_AUTH_TYPE, when set, always wins.
2. Otherwise bearer is used when BITBUCKET_OAUTH_TOKEN is set, or when no BITBUCKET_EMAIL is configured.
3. Otherwise basic is used (the default for personal API tokens).

Bearer example — a company-owned workspace access token with no personal account involved:

claude mcp add bitbucket -s user \
  -e BITBUCKET_WORKSPACE=your-workspace \
  -e BITBUCKET_OAUTH_TOKEN=your-workspace-access-token \
  -- npx mcp-server-bitbucket

Protecting the remote server (OAuth)

The stdio server runs on your machine with your own credential, so it needs no protection. A remote deployment is different: it holds one shared credential and serves whoever reaches it, so /mcp can be gated on a bearer token issued by an external authorization server (Microsoft Entra ID, Okta, Auth0…).

Protection is off unless both MCP_OAUTH_ISSUER and MCP_OAUTH_AUDIENCE are set, so existing deployments are unaffected. With them set, the server:

• answers unauthenticated /mcp requests with 401 and a WWW-Authenticate header pointing at /.well-known/oauth-protected-resource, which is what makes an MCP client start the OAuth flow;
• serves that metadata document (RFC 9728) naming the authorization server;
• verifies every request's JWT — signature against the issuer's JWKS, plus iss, aud and expiry — and returns 403 insufficient_scope when a valid token lacks the required scope;
• leaves /health open, so platform probes keep working.

Note what this does and does not do: it controls who may use the server. Calls still reach Bitbucket under the server's own credential, so it does not attribute actions to individual users.

Claude Code CLI

# TypeScript version
claude mcp add bitbucket -s user \
  -e BITBUCKET_WORKSPACE=your-workspace \
  -e BITBUCKET_EMAIL=your-email@example.com \
  -e BITBUCKET_API_TOKEN=your-api-token \
  -- npx mcp-server-bitbucket

# Python version
claude mcp add bitbucket -s user \
  -e BITBUCKET_WORKSPACE=your-workspace \
  -e BITBUCKET_EMAIL=your-email@example.com \
  -e BITBUCKET_API_TOKEN=your-api-token \
  -- mcp-server-bitbucket

Cursor IDE

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "bitbucket": {
      "command": "npx",
      "args": ["mcp-server-bitbucket"],
      "env": {
        "BITBUCKET_WORKSPACE": "your-workspace",
        "BITBUCKET_EMAIL": "your-email@example.com",
        "BITBUCKET_API_TOKEN": "your-api-token"
      }
    }
  }
}

Available Tools (58 total)

Repositories

Pull Requests

Pipelines

trigger_pipeline Parameters

Examples:

# Default pipeline on main
trigger_pipeline(repo_slug="my-repo")

# Custom pipeline
trigger_pipeline(repo_slug="my-repo", custom_pipeline="deploy-staging")

# Custom pipeline on specific commit with variables
trigger_pipeline(
    repo_slug="my-repo",
    commit="abc123def",
    custom_pipeline="deploy-prod",
    variables=[
        {"key": "ENV", "value": "production"},
        {"key": "SECRET", "value": "xxx", "secured": True}
    ]
)

Branches, Commits, Tags

And more...

• Deployments: list_environments, get_environment, list_deployment_history
• Webhooks: list_webhooks, create_webhook, get_webhook, delete_webhook
• Branch Restrictions: list_branch_restrictions, create_branch_restriction, delete_branch_restriction
• Source Browsing: get_file_content, list_directory
• Permissions: User and group permission management (8 tools)
• Projects: list_projects, get_project

MCP Prompts

Reusable workflow templates:

MCP Resources

Browsable workspace data:

Development

TypeScript

cd typescript
npm install
npm run build
npm run dev  # Watch mode

Python

cd python
uv sync
uv run python -m src.server

Creating a Bitbucket API Token

1. Go to your repository in Bitbucket
2. Navigate to Repository settings > Access tokens
3. Click Create Repository Access Token
4. Select permissions:
   • Repository: Read, Write, Admin, Delete
   • Pull requests: Read, Write
   • Pipelines: Read, Write
5. Copy the token immediately

Using this at work

If your team is running this against a company Bitbucket, I'd like to know. It decides what gets built next, and issues that come from a real deployment get looked at first.

Two kinds of work I take on around this:

• An MCP server for an internal system that doesn't have one. Same shape as this one, against your own API.
• Help getting an MCP deployment past a security review — scopes, token handling, self-hosting the transport.

Open an issue, or write to javier@javieraguilar.ai.

Author

Built and maintained by Javier Aguilar — AI engineer working on agent systems and MCP integrations. There's a write-up of how and why this server was built here.

License

MIT