Phabricator

A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact intelligently with Phabricator for advanced task management and code review workflows.

- Per-User Authentication: Configure your personal Phabricator API token in your MCP client
- User Attribution: Comments and reviews appear under YOUR name instead of a shared service account
- Flexible Configuration: Supports both personal tokens and shared environment variables
- Standard MCP Integration: Follows MCP ecosystem best practices for authentication

- Differential Management: View revisions, read comments, approve/reject code changes
- Intelligent Review Feedback: Analyze comments with surrounding code context for actionable insights
- Inline Comments: Add targeted feedback to specific lines in code reviews
- Code Context Analysis: Correlate review comments with actual code changes and locations

- HTTP/SSE Transport: FastMCP-based server for reliable production use (default on port 8932)
- stdio Transport: Legacy support for direct MCP client integration
- Comprehensive API: 11 specialized tools for complete Phabricator workflow automation

- Comment-Code Correlation: Intelligently link review feedback to specific code locations
- Contextual Code Display: Show surrounding code lines for better understanding
- Action Item Generation: Categorize feedback into actionable to-do items
- Priority Classification: Organize comments by Issues → Suggestions → Nits → Other

- Python 3.8+
- Phabricator instancewith API access
- API tokenfrom Phabricator (Settings → Conduit API Tokens)

# Clone and navigate git clone https://github.com/YushengAuggie/phabricator-mcp-server.git cd phabricator-mcp-server # Configure credentials echo "PHABRICATOR_TOKEN=your-32-character-api-token" > .env echo "PHABRICATOR_URL=https://your-phabricator-instance.com/api/" >> .env # Start server (handles all setup automatically) python3 start.py --mode http

The server starts onhttp://localhost:8932with automatic dependency management.

# Create virtual environment python3 -m venv venv source venv/bin/activate # Windows: venv\Scripts\activate # Install with dependencies pip install -e . # Start HTTP server python src/servers/http_server.py # Or start stdio server python src/servers/stdio_server.py

The server supportshybrid authenticationwith two modes that work seamlessly together:
- Personal API Token (Recommended): Pass your personal token through MCP client configuration for user attribution
- Environment Variable Fallback: Use a shared service account token via environment variables
- Go to your Phabricator instance → Settings → API Tokens
- Create a new token with appropriate permissions
- Copy the 32-character token for use in configuration

🌐 Finding Your Phabricator URL:Your Phabricator API URL should end with/api/and typically looks like:

- https://phabricator.example.com/api/
- https://phab.yourcompany.com/api/
- https://your-domain.phabricator.com/api/

If unsure, check your Phabricator instance's main page - the URL is usually[your-base-url]/api/

The server automatically detects your environment configuration:

claude mcp add --transport sse phabricator http://localhost:8932/sse \ --env "PHABRICATOR_TOKEN=api-xxxxxxx" \ --env "PHABRICATOR_URL=https://example.com/api/"

Replaceapi-xxxxxxxwith your actual API token andhttps://example.com/api/with your Phabricator instance URL

{ "mcpServers": { "phabricator": { "url": "http://localhost:8932/sse", "env": { "PHABRICATOR_TOKEN": "api-xxxxxxx", "PHABRICATOR_URL": "https://example.com/api/" } } } }

For Claude Desktop and direct MCP integration:

{ "mcpServers": { "phabricator": { "command": "python", "args": ["path/to/phabricator-mcp-server/start.py"], "cwd": "path/to/phabricator-mcp-server", "env": { "PHABRICATOR_TOKEN": "api-xxxxxxx", "PHABRICATOR_URL": "https://example.com/api/" } } } }

The server supports multiple ways to authenticate:
- Personal Token in Tools: Some tools accept anapi_tokenparameter
- Environment Variables: SetPHABRICATOR_TOKENin MCP client config
- Fallback Token: Create.envfile in server directory

Priority Order: Personal token → MCP environment → Server.envfile

Create.envfile in project root for fallback authentication:

# Fallback: Shared service account token PHABRICATOR_TOKEN=your-shared-token-here # Optional: Custom Phabricator URL (auto-detected from token by default) # PHABRICATOR_URL=https://your-phabricator-instance.com/api/ # Optional: Custom server port (default: 8932) # MCP_SERVER_PORT=8932

- Personal tokens: Comments appear under YOUR name
- Shared tokens: Comments appear under the service account name
- Mixed usage: Different tools can use different tokens

- Tokens are passed securely through MCP protocol
- No tokens stored on disk (except optional.envfallback)
- Each client can use their own personal token
- Check token validity: Test your token directly with Phabricator API
- Verify configuration: EnsurePHABRICATOR_TOKENis set correctly
- Check environment: Run server with debugging to see environment variables
- Use personal token: Passapi_tokenparameter directly to tools

# Check if server can start with your token PHABRICATOR_TOKEN=your-token python start.py --mode http # Test token manually curl -d "api.token=your-token" https://your-phabricator-instance.com/api/user.whoami

Add to Claude Desktop configuration (claude_desktop_config.json):

{ "mcpServers": { "phabricator": { "command": "python", "args": ["path/to/phabricator-mcp-server/start.py", "--mode", "stdio"], "cwd": "path/to/phabricator-mcp-server" } } }

{ "mcpServers": { "phabricator": { "url": "http://localhost:8932/sse" } } }

from src.core.client import PhabricatorClient # Initialize client client = PhabricatorClient( token="your-32-char-api-token", host="https://your-instance.com/api/" ) # Get enhanced review feedback with code context feedback = await client.get_review_feedback_with_code_context("12345", context_lines=7) # Add inline comment to specific line await client.add_inline_comment("12345", "src/file.py", 42, "Consider using a more descriptive variable name") # Get task with full context task = await client.get_task("6789") comments = await client.get_task_comments("6789")

# Get intelligent review feedback feedback_data = await client.get_review_feedback_with_code_context("D123", context_lines=5) # The feedback includes: # - Comments correlated with specific code locations # - Surrounding code context for each comment # - Action items categorized by priority # - File-by-file breakdown of changes

# Install with dev dependencies pip install -e ".[dev]" # Or with uv (faster) uv pip install -e ".[dev]"

# Run all tests with our test runner python run_tests.py # Run specific test suites python -m pytest src/tests/test_tool_completeness.py -v python -m pytest src/tests/test_tool_integration.py -v # Run with coverage python -m pytest --cov=src --cov-report=html

# Format code black src/ ruff check src/ --fix # Type checking mypy src/ # Run all quality checks black src/ && ruff check src/ && mypy src/ && python run_tests.py

- Tool Completeness: Validates all 11 tools are properly configured
- Integration Testing: Tests all tools with realistic mock data
- Error Handling: Validates graceful failure modes
- Argument Validation: Ensures correct required/optional parameters
- Mock Phabricator: No API calls needed for testing

Theget-review-feedbacktool provides advanced analysis:

# Returns structured feedback with: { "revision": {...}, # Revision metadata "review_feedback": [ # Enhanced comment analysis { "comment": "Fix this issue", "author": "reviewer-phid", "type": "inline", "code_context": { "file": "src/example.py", "target_line": 42, "hunk_info": "@@ -40,7 +40,7 @@", "lines": [ # Surrounding code context {"line_number": 40, "content": "def example():", "is_target": False}, {"line_number": 41, "content": " # TODO: fix this", "is_target": False}, {"line_number": 42, "content": " return broken_code", "is_target": True}, {"line_number": 43, "content": " # end function", "is_target": False}, ] }, "primary_file": "src/example.py", "primary_line": 42 } ], "summary": "Analysis summary with actionable insights", "total_comments": 5, "comments_with_context": 3 }

- Keyword Extraction: Identifies variable names, function names in comments
- Code Location Mapping: Links comments to specific files and line numbers
- Context Enrichment: Shows surrounding code for better understanding
- Priority Classification: Organizes feedback by importance

We welcome contributions! Here's how to get started:

# Fork and clone the repository git clone https://github.com/your-username/phabricator-mcp-server.git cd phabricator-mcp-server # Create feature branch git checkout -b feature/amazing-feature # Make changes and test python run_tests.py # Commit and push git commit -m 'feat: add amazing feature' git push origin feature/amazing-feature # Open a Pull Request

- Follow existing code style (black + ruff)
- Add tests for new features
- Update documentation as needed
- Ensure all quality checks pass

MIT License - seeLICENSEfile for details.

- Repository:https://github.com/YushengAuggie/phabricator-mcp-server
- Model Context Protocol:https://modelcontextprotocol.io/
- FastMCP:https://github.com/jlowin/fastmcp
- Phabricator API:https://secure.phabricator.com/book/phabricator/article/conduit/

A team task management application for collaboration between humans and AI.