This MCP server provides a Phabricator integration that enables AI assistants to manage tasks and perform code review actions directly against a Phabricator instance. It supports personal authentication via a user’s Phabricator API token as well as environment-based fallback tokens for shared service accounts. The server exposes 11 specialized tools across two domains: Task Management (view tasks, add comments, subscribe users) and Code Review (retrieve differential details, add comments, post inline feedback, and manage approvals). You can run the server with HTTP/SSE transport for production-grade usage or via stdio for direct MCP client integration. To use it, configure your MCP client with the target transport and supply authentication details (either personal token or environment-based token). The tools will return structured outputs including task metadata, comment threads, differential changes, and contextual code information to support AI-assisted workflows.

Prerequisites:

• Python 3.8+
• Access to a Phabricator instance with Conduit API enabled
• A Phabricator API token (personal or via environment fallback)

1. Clone the repository and navigate to the project root

   git clone https://github.com/YushengAuggie/phabricator-mcp-server.git cd phabricator-mcp-server

2. Create and activate a Python virtual environment

   python3 -m venv venv source venv/bin/activate # Unix/macOS

   Windows: venv\Scripts\activate

3. Install the package in editable mode (requirements are defined in setup.py / pyproject)

   pip install -e .

4. Prepare authentication configuration (example using environment variables or a .env file)

   Environment variable approach (preferred for runtime environments)

   export PHABRICATOR_TOKEN=api-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx export PHABRICATOR_URL=https://your-phabricator-instance.com/api/

   Optional: create a .env fallback in the server directory

   PHABRICATOR_TOKEN=api-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

   PHABRICATOR_URL=https://your-phabricator-instance.com/api/

5. Start the HTTP server (default port 8932) following the quickstart in the README

   python start.py --mode http

6. (Alternative) Start the stdio server for direct MCP client integration

   python start.py --mode stdio

7. Verify the server is reachable (for HTTP/SSE):

   curl http://localhost:8932/sse

Tips and caveats:

• Personal tokens will attribute actions (comments, reviews) to your user by default; environment tokens use the service account name.
• The server supports both HTTP/SSE transport and stdio transport; choose based on your deployment and client needs.
• If authentication fails, double-check PHABRICATOR_TOKEN and PHABRICATOR_URL values. The server can be started with debugging enabled to surface environment-related issues.
• For secure deployments, prefer environment-based tokens in a managed runtime (e.g., container or orchestrator) and avoid committing tokens to code or repos.
• The MCP client configuration can specify per-tool authentication preferences where supported (e.g., some tools may accept an api_token parameter directly).
• If you need to adjust the server port, set MCP_SERVER_PORT in the environment or .env fallback as documented in the README.
• When using the HTTP/SSE transport, the server listens by default on http://localhost:8932; ensure your firewall rules allow traffic on that port in production.
• For troubleshooting, test tokens directly against Phabricator’s API (e.g., user.whoami) to verify permissions before enabling in MCP.