Hatch

Introduction

Hatch is the package manager for managing Model Context Protocol (MCP) servers with environment isolation, multi-type dependency resolution, and multi-host deployment. Deploy MCP servers to Claude Desktop, VS Code, Cursor, Kiro, Codex, and other platforms with automatic dependency management.

The canonical documentation is at docs/index.md and published at https://hatch.readthedocs.io/en/latest/.

Key Features

• Environment Isolation — Create separate, isolated workspaces for different projects without conflicts
• Multi-Type Dependency Resolution — Automatically resolve and install system packages, Python packages, Docker containers, and Hatch packages
• Multi-Host Deployment — Configure MCP servers on multiple host platforms
• Package Validation — Ensure packages meet schema requirements before distribution
• Development-Focused — Optimized for rapid development and testing of MCP server ecosystems

Supported MCP Hosts

Hatch supports deployment to the following MCP host platforms:

• Claude Desktop — Anthropic's desktop application for Claude with native MCP support
• Claude Code — Claude integration for VS Code with MCP capabilities
• VS Code — Visual Studio Code with the MCP extension for tool integration
• Cursor — AI-first code editor with built-in MCP server support
• Kiro — Kiro IDE with MCP support
• Codex — OpenAI Codex with MCP server configuration support
• LM Studio — Local LLM inference platform with MCP server integration
• Google Gemini CLI — Command-line interface for Google's Gemini model with MCP support

Quick Start

Install from PyPI

pip install hatch-xclam

Verify installation:

hatch --version

Install from source

git clone https://github.com/CrackingShells/Hatch.git
cd Hatch
pip install -e .

Create your first environment and Hatch! MCP server package

# Create an isolated environment
hatch env create my_project

# Switch to it
hatch env use my_project

# Create a package template
hatch create my_mcp_server --description "My MCP server"

# Validate the package
hatch validate ./my_mcp_server

Deploy MCP servers to your tools

Package-First Deployment (Recommended) — Add a Hatch package and automatically configure it on Claude Desktop and Cursor:

hatch package add ./my_mcp_server --host claude-desktop,cursor

Direct Configuration (Advanced) — Configure arbitrary MCP servers on your hosts:

# Remote server example: GitHub MCP Server with authentication
export GIT_PAT_TOKEN=your_github_personal_access_token
hatch mcp configure github-mcp --host gemini \
  --httpUrl https://api.github.com/mcp \
  --header Authorization="Bearer $GIT_PAT_TOKEN"

# Local server example: Context7 via npx
hatch mcp configure context7 --host vscode \
  --command npx --args "-y @upstash/context7-mcp"

Documentation

• Full Documentation — Complete reference and guides
• Getting Started — Quick start for users
• CLI Reference — All commands and options
• Tutorials — Step-by-step guides from installation to package authoring
• MCP Host Configuration — Deploy to multiple platforms
• Developer Docs — Architecture, implementation guides, and contribution guidelines
• Troubleshooting — Common issues and solutions

Contributing

We welcome contributions! See the How to Contribute guide for details.

Quick start for developers

1. Fork and clone the repository
2. Install dependencies: pip install -e . and npm install
3. Create a feature branch: git checkout -b feat/your-feature
4. Make changes and add tests
5. Use conventional commits: npm run commit for guided commits
6. Run tests: wobble
7. Create a pull request

We use Conventional Commits for automated versioning. Use npm run commit for guided commit messages.

Getting Help

• Search existing GitHub Issues
• Read Troubleshooting for common problems
• Check Developer Onboarding for setup help

License

This project is licensed under the GNU Affero General Public License v3 — see LICENSE for details.