Phabricator

You can run a dedicated MCP server to securely connect your AI assistants to Phabricator for task management and code reviews. This server supports both HTTP/SSE transport for production use and a stdio transport for direct client integration, giving you flexible options to automate workflows with accurate attribution and rich context.

How to use

Start by choosing your transport method. For production-ready setups, run the HTTP/SSE transport so your MCP client talks to the server over a fast, reliable connection. You can also run a stdio transport for direct MCP integration from your local environment. Use the provided credentials and Phabricator URL to authorize and attribute actions to your user when desired.

Once the server is running, your MCP client can perform all supported operations: fetch task details, read and add comments, subscribe users to tasks, view differential revisions, read and attach inline comments, and approve or request changes on code reviews. The server analyzes review feedback with surrounding code context, links comments to exact code locations, and groups feedback into actionable items with clear priority categories.

How to install

Prerequisites: Python 3.8 or newer and a Phabricator instance with API access.

Automated setup (recommended): clone, configure credentials, and start the server in HTTP mode.

Configuration and usage notes

Authentication can be done via a personal Phabricator API token for user attribution or via environment variables for a shared service account. You can mix methods, but tokens are applied per tool, and personal tokens ensure actions appear under your name.

Key environment variables you’ll configure in your MCP client or environment include the Phabricator token and the API URL (ending with /api/). These values enable secure requests and proper attribution.

For HTTP/SSE transport, your client connects to the server at the SSE endpoint and passes in tokens and URLs as part of the environment. For stdio transport, the server is started locally and the client communicates directly via standard input/output.

If you encounter authentication issues, verify token validity, correct environment variable names, and the server startup mode. Prefer using a personal token when possible for clear attribution.

Troubleshooting tips

Check that the server is reachable at the expected port (default 8932) and that the SSE endpoint is responding. Validate tokens by performing a simple API call to the Phabricator Conduit API. If errors persist, run the server with debugging enabled to inspect environment variables and startup messages.

Available tools

get-task

Fetch comprehensive task details including metadata, status, and full comment threads.

add-task-comment

Post a comment to a specific Phabricator task to facilitate collaboration.

subscribe-to-task

Subscribe users to task notifications to stay informed about updates.

get-differential

Retrieve basic information about a differential revision.

get-differential-detailed

Fetch a detailed view of a revision including code changes.

get-review-feedback

Analyze review comments with surrounding code context and provide actionable insights.

add-differential-comment

Add general comments to a differential revision.

add-inline-comment

Attach targeted comments to specific lines in code changes.

accept-differential

Approve a differential revision after review.

request-changes-differential

Request changes on a differential revision with optional feedback.

subscribe-to-differential

Subscribe users to differential notifications