Install MCP 🚀

The simplest way to manage MCP servers through your AI agent

Install MCP is a meta-MCP server that enables AI agents to install, configure, and manage other MCP servers for you. No more manual JSON editing or complex command-line installations - just ask your AI assistant to handle it.

🌟 Key Features

• Agent-First Design: Install and manage MCP servers by simply asking your AI
• Universal Compatibility: Works with Claude Desktop, Claude Code, Cursor, and Windsurf
• Centralized Management: All configurations and secrets stored in one place
• Secure Secret Handling: Web-based secure collection for API keys and tokens
• One-Line Installation: Get started in seconds with a simple command
• Export/Import Setups: Share your MCP configuration with your team

📦 Installation

Our installer automatically sets up uv (the fast Python package manager) and configures everything for you.

macOS/Linux

curl -sSL https://raw.githubusercontent.com/ddfourtwo/install-mcp/main/install.sh | bash

Windows

powershell -c "irm https://raw.githubusercontent.com/ddfourtwo/install-mcp/main/install.ps1 | iex"

What the installer does:

• ✅ Installs uv if not present (blazing fast Python package manager)
• ✅ Adds uv to your PATH permanently
• ✅ Downloads and configures install-mcp
• ✅ Sets up the server to use uvx for instant startup
• ✅ Configures all your MCP clients automatically

After installation:

1. Restart your MCP client (Claude Desktop, Cursor, etc.)
2. Ask your AI: "Can you test the install-mcp server?"
3. You're ready to install any MCP server!

💬 Usage Examples

Once installed, just ask your AI assistant:

Installing Servers

"Install the GitHub MCP server for me"
"Can you set up the Slack MCP server?"
"I need the filesystem MCP server installed"

Managing Configurations

"Show me all my installed MCP servers"
"Export my MCP setup so I can share it"
"Configure the GitHub server in all my MCP clients"

Handling Secrets

"Collect the API key for the OpenAI server"
"Update the GitHub token for my GitHub MCP server"

📖 Documentation

• Architecture Guide - Technical overview with diagrams
• API Reference - Detailed documentation of all available tools

🚀 Why uv?

We use uv as our Python package manager because it's:

• 10-100x faster than pip - installs complete in seconds, not minutes
• Universal - single binary that works everywhere
• Modern - built in Rust with proper dependency resolution
• Convenient - uvx works just like npx for running Python tools

After our installer runs, you can use uvx anywhere:

# Run any Python tool instantly
uvx ruff check
uvx black --version
uvx install-mcp  # This server!

🛠️ How It Works

Install MCP acts as a package manager for MCP servers:

1. Installation: Handles git cloning, npm installs, and dependency management
2. Configuration: Maintains a central config file that syncs to all MCP clients
3. Secret Management: Securely collects and stores API keys in a central .env file
4. Client Updates: Automatically updates Claude Desktop, Cursor, and other MCP clients

Directory Structure

~/mcp-servers/
├── install-mcp/          # This meta-server
├── mcp-servers-config.json  # Central configuration
├── .env                  # Central secrets (secure)
├── github/              # Example: GitHub MCP server
├── slack/               # Example: Slack MCP server
└── ...                  # Other installed servers

🔧 Available AI Commands

Your AI assistant can use these tools:

📚 Supported Server Types

Install MCP can handle various MCP server types:

NPM Packages

# Servers published to npm
Command: npx -y @organization/package-name@latest

GitHub Repositories (Node.js)

# Servers with package.json
1. Clone repository
2. Run npm install
3. Run npm build (if needed)
4. Configure with: node path/to/server.js

GitHub Repositories (Python)

# Servers with requirements.txt
1. Clone repository  
2. Run pip install -r requirements.txt
3. Configure with: python3 path/to/server.py

Pre-built Executables

# Standalone binaries
Configure with direct path to executable

🔐 Security

• Secrets are never exposed in the AI conversation
• API keys are collected via a local-only web interface
• Secrets are stored with restrictive permissions (0600)
• Each server's secrets are namespaced (e.g., GITHUB_TOKEN → GITHUB_GITHUB_TOKEN)

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Setup

# Clone the repository
git clone https://github.com/ddfourtwo/install-mcp.git
cd install-mcp

# Install dependencies
cd server
pip install -r requirements.txt

Running Tests

python test_meta_mcp.py

📋 Troubleshooting

MCP client not recognizing the server?

• Ensure you've restarted your MCP client after installation
• Check that the install path is correct: ~/mcp-servers/install-mcp/
• Verify Python 3.11+ is installed: python3 --version

Secrets not working?

• Check the central .env file: ~/mcp-servers/.env
• Ensure secrets are properly namespaced (e.g., GITHUB_GITHUB_TOKEN)
• Verify file permissions: ls -la ~/mcp-servers/.env (should be -rw-------)

Can't install a specific server?

• Run list_mcp_servers() to check current status
• Ensure you have the required tools (git, npm, python3)
• Check the server's specific requirements in its README

🌐 Ecosystem

Install MCP is part of the growing MCP (Model Context Protocol) ecosystem:

• MCP Specification
• Awesome MCP Servers
• Claude Desktop

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• Built with FastMCP framework
• Inspired by the need for simpler MCP server management
• Thanks to all contributors and early adopters

Made with ❤️ for the AI-assisted development community

Remember: The best interface for AI tools is natural language. Let your AI assistant handle the complexity!