English | 简体中文

Open Source. Forever Free.
Built with ❤️ by Path Integral Institute

MCPM is an open source CLI tool for managing MCP servers. It provides a simplified global configuration approach where you install servers once and organize them with profiles, then integrate them into any MCP client. Features include server discovery through a central registry, direct execution, sharing capabilities, and client integration tools.

Demo is showing outdated v1 commands, new demo is baking ...

💡 Grow the MCP ecosystem! We welcome contributions to our MCP Registry. Add your own servers, improve documentation, or suggest features. Open source thrives with community participation!

curl -sSL https://mcpm.sh/install | bash

Or choose other installation methods like brew, pipx, uv etc.

MCPM v2.0 provides a simplified approach to managing MCP servers with a global configuration model. Key features include:

• ✨ Global Server Management: Install servers once, use everywhere
• 📋 Virtual Profiles: Organize servers with tags for different workflows
• 🔍 Server Discovery: Browse and install from the MCP Registry
• 🚀 Direct Execution: Run servers over stdio or HTTP for testing
• 🌐 Public Sharing: Share servers through secure tunnels
• 🔄 Server Updates: Check for and apply updates to git-based, NPX, and binary servers
• 🎛️ Client Integration: Manage configurations for Claude Desktop, Cursor, Windsurf, and more
• 🤖 AI Agent Friendly: Non-interactive CLI with comprehensive automation support and llm.txt guide
• 💻 Beautiful CLI: Rich formatting and interactive interfaces
• 📊 Usage Analytics: Monitor server usage and performance

MCPM v2.0 eliminates the complexity of v1's target-based system in favor of a clean global workspace model.

MCPM will support managing MCP servers for the following clients:

• 🤖 Claude Desktop (Anthropic)
• ⌨️ Cursor
• 🏄 Windsurf
• 🧩 Vscode
• 📝 Cline
• ➡️ Continue
• 🦢 Goose
• 🔥 5ire
• 🦘 Roo Code
• 💻 OpenCode
• ✨ More clients coming soon...

MCPM provides a comprehensive CLI with a clean, organized interface. The v2.0 architecture uses a global configuration model where servers are installed once and can be organized with profiles, then integrated into specific MCP clients as needed.

mcpm --help          # Display help information and available commands
mcpm --version       # Display the current version of MCPM

Global server installation and management commands:

# 🔍 Search and Install
mcpm search [QUERY]           # Search the MCP Registry for available servers
mcpm info SERVER_NAME         # Display detailed information about a server
mcpm install SERVER_NAME      # Install a server from registry to global configuration
mcpm uninstall SERVER_NAME    # Remove a server from global configuration

# 📋 List and Inspect
mcpm ls                       # List all installed servers and their profile assignments
mcpm edit SERVER_NAME         # Edit a server configuration
mcpm inspect SERVER_NAME      # Launch MCP Inspector to test/debug a server

Execute servers directly for testing or integration:

mcpm run SERVER_NAME          # Execute a server directly over stdio
mcpm run SERVER_NAME --http   # Execute a server over HTTP for testing
mcpm share SERVER_NAME        # Share a server through secure tunnel for remote access
mcpm usage                    # Display comprehensive analytics and usage data

Profiles are virtual tags that organize servers into logical groups for different workflows:

# 🔄 Profile Operations
mcpm profile ls               # List all profiles and their tagged servers
mcpm profile create PROFILE   # Create a new profile
mcpm profile rm PROFILE       # Remove a profile (servers remain installed)
mcpm profile edit PROFILE     # Interactive server selection for profile

# 🚀 Profile Execution
mcpm profile run PROFILE      # Execute all servers in a profile over stdio or HTTP
mcpm profile share PROFILE    # Share all servers in a profile through secure tunnel
mcpm profile inspect PROFILE  # Launch MCP Inspector for all servers in profile

Manage MCP client configurations (Claude Desktop, Cursor, Windsurf, etc.):

mcpm client ls                 # List all supported MCP clients and their status
mcpm client edit CLIENT_NAME   # Interactive server enable/disable for a client
mcpm client edit CLIENT_NAME -e # Open client config in external editor
mcpm client import CLIENT_NAME  # Import server configurations from a client

Check for and apply updates to installed MCP servers:

mcpm update                   # Update all servers
mcpm update SERVER_NAME       # Update a specific server
mcpm update --check           # Dry run — check for updates without applying
mcpm update --rebase          # Use git rebase instead of fast-forward
mcpm update --init            # Scan servers and populate source metadata
mcpm update --init --force    # Re-detect all source metadata

MCPM tracks where each server came from (git repo, npm package, HTTP remote) and can pull the latest changes automatically. Git-based servers are updated via git pull --ff-only by default, with --rebase as an opt-in alternative. NPX/UVX servers auto-update at runtime and are shown for informational purposes.

mcpm doctor                   # Check system health and server status
mcpm config                   # Manage MCPM configuration and settings
mcpm migrate                  # Migrate from v1 to v2 configuration

The MCP Registry is a central repository of available MCP servers that can be installed using MCPM. The registry is available at mcpm.sh/registry.

MCPM is designed to be AI agent friendly with comprehensive automation support. Every interactive command has a non-interactive alternative using CLI parameters and environment variables.

Set environment variables to enable full automation:

export MCPM_NON_INTERACTIVE=true  # Disable all interactive prompts
export MCPM_FORCE=true            # Skip confirmations
export MCPM_JSON_OUTPUT=true      # JSON output for parsing

The llm.txt file provides comprehensive documentation specifically designed for AI agents, including:

• Complete command reference with parameters and examples
• Environment variable usage patterns
• Best practices for automation
• Error handling and troubleshooting
• Batch operation patterns

The llm.txt file is automatically generated from the CLI structure and kept up-to-date with each release.

1. Add the server to MCPM

mcpm new [MY_SERVER_NAME] --type stdio --command "[COMMAND]" --args "[ARG_1] [ARG_2] ..."

(Add --force arg to bypass the confirmation interaction) Verify it appears in your installed servers:

mcpm ls

You should see [MY_SERVER_NAME] in the list.

2. Serve it over HTTP

mcpm run [MY_SERVER_NAME] --http

MCPM prints the server URL, for example:

URL: http://127.0.0.1:6276/mcp/

3. Expose it on your network (optional)

mcpm run [MY_SERVER_NAME] --http --host 0.0.0.0

Tip: add --port <PORT> to choose a different port.

1. Add the servers to MCPM

# Stdio server
mcpm new [SERVER_A] --type stdio --command "[COMMAND]" --args "[ARG_1] [ARG_2] ..." --force

# Remote HTTP/SSE server
mcpm new [SERVER_B] --type remote --url "http://SERVER_ADDR/mcp/" --force

2. Create a profile and add the servers Interactive:

mcpm profile create [PROFILE_NAME]
mcpm profile edit [PROFILE_NAME]

MCPM would list out all installed servers, you can use arrows to navigate, space to select/deselect, type to search, and Enter to confirm.

Verify:

mcpm profile ls

You should see your profile with the selected servers.

3. Run the profile

# Stdio
mcpm profile run [PROFILE_NAME]

# HTTP
mcpm profile run [PROFILE_NAME] --http

# Optional: host and port
mcpm profile run [PROFILE_NAME] --http --host 0.0.0.0 --port 8080

# Server management
mcpm new myserver --type stdio --command "python -m server" --force
mcpm edit myserver --env "API_KEY=secret" --force

# Profile management  
mcpm profile edit web-dev --add-server myserver --force
mcpm profile run web-dev --port 8080

# Client integration
mcpm client edit cursor --add-profile web-dev --force

• Global server configuration model
• Profile-based server tagging and organization
• Interactive command interfaces
• Client integration management (mcpm client edit)
• Modern CLI with consistent UX
• Registry integration and server discovery
• Direct server execution and sharing
• Import from existing client configurations
• Additional client support (gemini-cli, codex, etc.)

• Server update management (mcpm update)
• GitHub release binary updates (download + replace)
• Advanced Server access monitoring and analytics
• Execution in docker
• Expose MCPM functionality as an MCP server (search, install, profile management etc.)
• TUI interface for MCP inspect
• TUI interface for MCP & profile management

brew install mcpm

pipx install mcpm

uv tool install mcpm

pip install mcpm

If you are a user of x-cmd, you can run:

x install mcpm.sh

This repository contains the CLI and service components for MCP Manager, built with Python and Click following modern package development practices.

• 🐍 Python 3.10+
• 🚀 uv (for virtual environment and dependency management)
• 🖱️ Click framework for CLI
• ✨ Rich for enhanced console output
• 🌐 Requests for API interactions

The project follows the modern src-based layout:

mcpm.sh/
├── src/             # Source package directory
│   └── mcpm/        # Main package code
├── tests/           # Test directory
├── test_cli.py      # Development CLI runner
├── pyproject.toml   # Project configuration
├── pages/           # Website content
│   └── registry/    # Registry website
├── mcp-registry/    # MCP Registry data
└── README.md        # Documentation

1. Clone the repository

   git clone https://github.com/pathintegral-institute/mcpm.sh.git
   cd mcpm.sh
   

2. Set up a virtual environment with uv

   uv venv --seed
   source .venv/bin/activate  # On Unix/Mac
   

3. Install dependencies in development mode

   uv pip install -e .
   

4. Run the CLI directly during development

   # Either use the installed package
   mcpm --help
   
   # Or use the development script
   ./test_cli.py --help
   

5. Run tests

   pytest tests/
   

• 📁 Use the src-based directory structure to prevent import confusion
• 🔧 Develop with an editable install using uv pip install -e .
• 🧩 Keep commands modular in the src/mcpm/commands/ directory
• 🧪 Add tests for new functionality in the tests/ directory
• 💻 Use the test_cli.py script for quick development testing

MCP uses a single source of truth pattern for version management to ensure consistency across all components.

• 📍 The canonical version is defined in version.py at the project root
• 📥 src/mcpm/__init__.py imports this version
• 📄 pyproject.toml uses dynamic versioning to read from version.py
• 🏷️ Git tags are created with the same version number prefixed with 'v' (e.g., v1.0.0)

When releasing a new version:

1. Use the provided version bump script

   ./bump_version.sh NEW_VERSION
   # Example: ./bump_version.sh 1.1.0
   

2. Push the changes and tags

   git push && git push --tags
   

3. Create a GitHub release matching the new version

This process ensures that the version is consistent in all places: code, package metadata, and git tags. PyPI release is handled by the CI/CD pipeline and will be triggered automatically.

MIT