phabricator

This MCP server wraps Phabricator's Conduit API, exposing Phabricator resources such as tasks (Maniphest), code reviews (Differential), repositories (Diffusion), and more to any MCP client. Once running, clients can query and operate on Phabricator data through the MCP abstraction layer, enabling tools to search, create, edit, and read resources without needing direct Conduit integration. The server uses environment configuration or your existing Arcanist setup to authorize requests against your Phabricator instance. Typical use cases include: listing tasks, inspecting revisions, creating or updating tasks, and retrieving repository details via a consistent MCP interface. Tools available in the MCP surface map to Phabricator concepts such as phabricator_task_search, phabricator_revision_search, phabricator_repository_search, and many related read/write operations. To get started, configure PHABRICATOR_URL and PHABRICATOR_API_TOKEN, then run the MCP server with the recommended command, and connect your MCP client to the server named phabricator.

Prerequisites:

• Node.js and npm (or a Node.js runtime that can run npx)
• Access to a Phabricator instance with a Conduit API token

Installation steps:

1. Install and configure prerequisites (Node.js and npm):

   • On macOS: install via Homebrew: brew install node
   • On Windows: install from nodejs.org
   • On Linux: use your distribution package manager or nvm to install Node.js

2. Install the MCP server using npx (as described in the README):

   claude mcp add --scope user phabricator -- npx @freelancercom/phabricator-mcp@latest

   Or, if you’re configuring via Codex/OpenAI/other tools, prefer the JSON-based config shown below:

   { "mcpServers": { "phabricator": { "command": "npx", "args": ["@freelancercom/phabricator-mcp@latest"], "env": { "PHABRICATOR_URL": "https://phabricator.example.com", "PHABRICATOR_API_TOKEN": "api-xxxxxxxxxxxxx" } } } }

3. Ensure environment variables are set if you’re not using a global arc configuration:

• PHABRICATOR_URL: URL of your Phabricator instance
• PHABRICATOR_API_TOKEN: Conduit API token for authentication

4. Start or configure your MCP client to connect to the phabricator server as shown in the examples.

5. If you need to upgrade, re-run the installation command to pick up the latest version.

Tips and notes:

• PHABRICATOR_URL and PHABRICATOR_API_TOKEN can be supplied via environment variables or your MCP client’s integration config. Environment variables take precedence over any local .arcrc configuration.
• The MCP server relies on your Phabricator Conduit API token scope. If you encounter permission prompts or limited access, ensure the token has the required Conduit capabilities for the tools you intend to use.
• To avoid repeatedly fetching new versions, you can pin a specific version tag in the npm argument (e.g., @freelancercom/phabricator-mcp@1.0.0). If you do pin, remember to periodically update and clear caches as needed.
• The provided tools set includes searching tasks, revisions, repositories, and more. If you want read-only operation without prompts, consider enabling the read-only tool permissions by whitelisting the mcp__phabricator__* tools in your Claude/CLI settings as described in the README.
• If you run into network or rate-limiting issues, verify that your Conduit API tokens are valid and that your Phabricator instance allows API access from your host.
• When migrating from GitHub-hosted configurations to the npm package, update your MCP client configs to reference the @freelancercom/phabricator-mcp package and adjust any JSON args accordingly.