mcp-pyrefly-autotype

lolpack/mcp-pyrefly-autotype

3.2

If you are the rightful owner of mcp-pyrefly-autotype and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

The MCP Pyrefly Autotype Server is a Model Context Protocol server designed to provide automatic Python type annotation using Pyrefly, enabling seamless integration with LLMs and AI coding assistants.

Tools
4
Resources
0
Prompts
0

[WORK IN PROGRESS AND UNTESTED - USE AT OWN RISK] MCP Pyrefly Autotype Server

A Model Context Protocol (MCP) server that provides automatic Python type annotation using Pyrefly. This server enables LLMs and AI coding assistants to analyze Python code, add type annotations, and perform type checking seamlessly.

What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard that enables AI assistants and language models to securely access external data sources and tools. MCP servers act as bridges between AI systems and various resources, providing structured access to information and capabilities.

How MCP Works

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   LLM/AI Client │◄──►│   MCP Server    │◄──►│  External Tools │
│   (e.g. Claude) │    │  (This Project) │    │   (Pyrefly)     │
└─────────────────┘    └─────────────────┘    └─────────────────┘

MCP servers can provide:

  • Resources: Static or dynamic data sources (files, databases, APIs)
  • Tools: Executable functions that perform actions
  • Prompts: Templated prompts for specific tasks

This allows AI assistants to:

  • Access real-time information
  • Perform complex operations
  • Integrate with existing tools and workflows
  • Maintain security through controlled access

Features

This MCP server provides comprehensive Python type annotation capabilities:

🔍 Analysis Tools

  • File Analysis: Analyze individual Python files for missing type annotations
  • Project Context: Get project-wide type information for better inference
  • Pyrefly Integration: Leverage Pyrefly's powerful type inference engine

⚡ Type Enhancement

  • Automatic Type Addition: Add type annotations using Pyrefly's autotype feature
  • File-based Processing: Process individual Python files with type annotations
  • Optional Backup: Can create backup files before modification (when requested)
  • Project Integration: Respects pyrefly configuration files

✅ Type Checking

  • Pyrefly Integration: Validate type annotations using Pyrefly's built-in type checker
  • Error Reporting: Basic type checking results and error output
  • File-based Validation: Check individual files for type errors

🤖 LLM Integration

  • Basic Prompts: Pre-built prompts for type analysis tasks
  • Structured Data: JSON-formatted analysis results
  • Simple Workflows: Basic analyze → annotate → verify workflows

Why Use This MCP Server?

For LLMs and AI Assistants

  • MCP Integration: Works with MCP-compatible AI clients
  • JSON Responses: Provides structured data for better decision making
  • Basic Context: Simple project structure analysis
  • Error Handling: Basic error reporting and graceful failure handling

For Developers

  • Cold Start Helper: Assists with completely untyped codebases
  • Basic Typing: Simple type annotation workflows
  • File Processing: Individual file type checking and annotation
  • Tool Integration: Basic integration with existing Python development workflows

Installation

Prerequisites

Install the MCP Server

# Clone or download this repository
git clone https://github.com/your-username/mcp-pyrefly-autotype.git
cd mcp-pyrefly-autotype

# Install dependencies with uv
uv sync

# For development (includes dev dependencies)
uv sync --dev

# Alternative: traditional pip install
# pip install -e .
# pip install -e ".[dev]"

Usage

Running the Server

The server can be run directly or integrated with MCP-compatible clients:

# Run directly (for testing)
uv run python -m mcp_pyrefly_autotype.server

# Or use the installed script (after uv sync)
uv run mcp-pyrefly-autotype

# Alternative: activate virtual environment first
uv shell
python -m mcp_pyrefly_autotype.server

Integration with AI Clients

Claude Desktop (Example Configuration)

Add to your Claude Desktop configuration:

{
  "mcpServers": {
    "pyrefly-autotype": {
      "command": "uv",
      "args": ["run", "python", "-m", "mcp_pyrefly_autotype.server"],
      "env": {}
    }
  }
}
VS Code with Copilot
  1. Install the MCP extension for VS Code
  2. Configure the server in your workspace settings:
  3. Create a .vscode/mcp.json file
{
	"servers": {
		"pyrefly-autotype": {
			"type": "stdio",
			"command": "uv",
			"args": [
				"run",
				"mcp-pyrefly-autotype"
			]
		}
	},
	"inputs": []
}
Make it show up in VS Code (MCP Servers + Copilot Chat Tools)
  1. Install the “Model Context Protocol (MCP)” extension in VS Code and ensure GitHub Copilot is enabled/updated.
  2. Save the .vscode/mcp.json file shown above in the root of your workspace.
  3. Reload the window: press Ctrl+Shift+P → “Developer: Reload Window”.
  4. Verify in the MCP Servers view:
    • Open the Command Palette (Ctrl+Shift+P) → run “MCP: Show Servers”, or open the “MCP Servers” view from the Activity Bar.
    • You should see a server named pyrefly-autotype. Status should be Running. If not:
      • Confirm uv is installed and on PATH, and that uv sync has been run.
      • On Windows, you may need to restart VS Code after installing Python/uv.
  5. Verify in Copilot Chat Tools:
    • Open Copilot Chat (Ctrl+I or the Copilot icon).
    • In the Tools pane, expand the MCP section. You should see pyrefly-autotype listed. If it’s missing, check that:
      • The workspace is trusted (look for the “Trust” banner in VS Code).
      • MCP integration is enabled in Copilot settings.
Run sample queries (inside Copilot Chat)

Try these prompts in a new Copilot Chat tab. Copilot will call the server’s tools for you.

  • “Use the pyrefly-autotype MCP server to analyze the file simple_untyped.py (detailed=true), then add types to it, and finally type check it. Repeat add→check up to 3 times until type check passes.”
  • “Analyze example_untyped.py for missing annotations, add types with a backup, and run a type check. Summarize changes and remaining warnings.”
  • “Given the loop in SamplePrompt.md, run the agent loop on simple_untyped.py: add_types_to_file → type_check_file, refining up to 3 rounds.”

Expected outcomes:

  • Copilot will invoke these MCP tools: analyze_python_file, add_types_to_file, type_check_file.
  • The file will be annotated in-place (a backup may be created depending on your request).
  • You’ll receive a summary and any remaining non-blocking warnings.

Available Tools

analyze_python_file

Analyze a Python file for missing type annotations.

Parameters:

  • file_path (required): Path to the Python file
  • detailed (optional): Include detailed analysis information

Example:

# LLM can request:
# "Analyze the file 'src/utils.py' for type annotation needs"

add_types_to_file

Add type annotations to a Python file using Pyrefly (this invokes pyrefly autotype under the hood).

Parameters:

  • file_path (required): Path to the Python file
  • backup (optional): Create backup before modifying (default: true)

Example:

# LLM can request:
# "Add type annotations to 'src/models.py'"

type_check_file

Run type checking on a Python file using Pyrefly.

Parameters:

  • file_path (required): Path to the Python file

Example:

# LLM can request:
# "Type check the file 'src/api.py' and report any errors"

get_project_context

Get project-wide type information for better inference.

Parameters:

  • project_path (required): Path to the project directory

Example:

# LLM can request:
# "Analyze the project structure for type annotation opportunities"

Available Prompts

analyze_typing_needs

Generate analysis prompts for type annotation needs.

type_improvement_plan

Create a comprehensive plan for improving type coverage in a project.

Example Workflows

1. Complete File Type Enhancement

# LLM workflow:
# 1. "Analyze 'calculator.py' for type needs"
# 2. "Add types to 'calculator.py'"
# 3. "Type check 'calculator.py' and report results"

2. Project-Wide Type Analysis

# LLM workflow:
# 1. "Get project context for '/my-project'"
# 2. "Create a type improvement plan for the project"
# 3. "Prioritize files for type annotation"

3. Cold Start Type Addition

# For completely untyped files:
# 1. "Analyze 'legacy_code.py' - it has no types at all"
# 2. "Add types to 'legacy_code.py'" 
# 3. "Check for type errors and suggest corrections"

Use Cases

🥶 Cold Start Projects

  • Challenge: Legacy codebases with no type annotations
  • Solution: Use Pyrefly autotype with basic MCP integration
  • Benefit: Start adding types to untyped codebases

📈 Incremental Typing

  • Challenge: Adding types to active projects gradually
  • Solution: File-by-file type annotation with basic project context
  • Benefit: Gradual type adoption without major disruption

🔧 CI/CD Integration

  • Challenge: Maintaining type quality in team projects
  • Solution: Basic type checking integration in pipelines
  • Benefit: Simple type validation workflows

🤝 LLM-Assisted Development

  • Challenge: LLMs need context about typing needs
  • Solution: Basic structured analysis data and simple prompts
  • Benefit: Improved AI assistance for Python type annotation tasks

Configuration

Pyrefly Configuration

The server respects Pyrefly's configuration. You can configure Pyrefly in your project using either:

  1. pyrefly.toml file in your project root:
# Files to include in type checking  
project-includes = ["src/**/*.py"]

# Files to exclude from type checking
project-excludes = ["tests/**", "**/__pycache__/**"]

# Python version to assume
python-version = "3.12"

# How to handle untyped function definitions
untyped-def-behavior = "check-and-infer-return-type"

# Configure specific error types
[errors]
# Enable/disable specific error types
bad-assignment = true
missing-return-type = true
  1. pyproject.toml file under the [tool.pyrefly] section:
[tool.pyrefly]
# Files to include in type checking
project-includes = ["src/**/*.py"]

# Files to exclude from type checking  
project-excludes = ["tests/**", "**/__pycache__/**"]

# Python version and platform
python-version = "3.12"
python-platform = "linux"

# Type checking behavior
untyped-def-behavior = "check-and-infer-return-type"
ignore-missing-imports = ["requests.*", "numpy.*"]

# Error configuration
[tool.pyrefly.errors]
bad-assignment = true
missing-return-type = true

See the Pyrefly Configuration Documentation for all available options.

Development

Running Tests

# Run all tests
uv run pytest tests/

# Run with coverage
uv run pytest tests/ --cov=mcp_pyrefly_autotype

# Run specific test
uv run python tests/test_server.py

# Test server functions directly
uv run python test_direct.py

# Run demo workflow
uv run python test_demo.py

Testing the MCP Server

The project includes several test files to verify functionality:

  • tests/test_server.py - Comprehensive test suite with mocked pyrefly calls
  • test_direct.py - Direct testing of server functions with real pyrefly
  • test_demo.py - Interactive demo showing the complete workflow
  • simple_untyped.py - Example file for testing type annotation

To test the server end-to-end:

# 1. Test with a simple untyped file
uv run python test_demo.py

# 2. Test server functions directly  
uv run python test_direct.py

# 3. Run the MCP server (for client integration)
uv run python -m mcp_pyrefly_autotype.server

Code Quality

# Format code
uv run black src/ tests/

# Lint code  
uv run ruff check src/ tests/

# Type check
uv run pyrefly check src/

Contributing

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes (git commit -m 'Add amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

License

This project is licensed under the MIT License - see the file for details.

Related Projects

Support

For questions and support:

  • Open an issue on GitHub
  • Check the Pyrefly documentation
  • Review the MCP specification

This MCP server bridges the gap between AI assistants and Python type annotation tools, enabling seamless integration of type enhancement workflows in AI-powered development environments.

Sample Queries and Prompt Library

The sample_queries/ directory contains ready-to-use prompt templates you can paste into your AI client (VS Code with Copilot MCP or Claude Desktop) to drive the server effectively:

  • sample_queries/PromptWithTools.md — A compact “agent loop” prompt that instructs the assistant to use the available tools (add_types_to_file, type_check_file, and optionally get_project_context) and iterate up to 3 refinement rounds. Great for single-file or small feature work.
  • sample_queries/LargeUntypedCodebase.md — A batch-oriented workflow for incrementally typing a large, mostly-untyped repo. It includes planning, per-file refine loops, batch gates, and progress tracking guidance.

How to use with VS Code + Copilot:

  • Open Copilot Chat. Ensure the pyrefly-autotype MCP server appears under Tools (see instructions above).
  • Open one of the markdown files in sample_queries/, copy the prompt, and paste it into Copilot Chat.
  • If the prompt includes tool call JSON examples, Copilot will translate them into MCP tool invocations automatically.

How to use with Claude Desktop:

  • Ensure your Claude MCP configuration includes this server (see “Claude Desktop” section above).
  • Open a new chat, paste any of the prompts, and follow the agent’s steps. Claude will call the MCP tools using the provided JSON shapes.

Tip: Start with PromptWithTools.md on a single file (e.g., simple_untyped.py) to see the full add → check → refine flow end to end. Then progress to LargeUntypedCodebase.md for multi-file, incremental adoption.