MCP Anywhere acts as a unified gateway for MCP servers, providing discovery, credential management, and secure, containerized execution of tools sourced from GitHub repositories. Through a web interface and programmatic API, you can add MCP server repositories, enable or disable individual tools, and manage credentials and secret files. The system uses Claude AI to analyze and configure repositories automatically, accelerating onboarding of new tools. You can also run tools via a CLI or Claude Desktop integration, and leverage Docker containerization to isolate tool execution for security and reproducibility. The gateway consolidates API keys, secrets, and access controls, offering a single endpoint to access a growing catalog of MCP tools and runtimes, including Python interpreters, npx, uvx, and Docker-based runtimes.

Prerequisites:

• Git
• Python 3.8+ (recommended) with pip
• Optional: uv (if you plan to install via uv) or Docker for containerized deployment

Installation steps:

1. Clone the repository

git clone https://github.com/locomotive-agency/mcp-anywhere.git
cd mcp-anywhere

2. Install dependencies (recommended via uv) or via pip Option A — using uv (recommended):

uv sync

Option B — via pip (editable install):

pip install -e .

3. Configure environment Copy the example environment file and edit with required values:

cp env.example .env

Edit .env to include at least:

• SECRET_KEY: a secure random key
• ANTHROPIC_API_KEY: your API key (if using Claude AI features)
• Optional: GOOGLE_OAUTH_CLIENT_ID, GOOGLE_OAUTH_CLIENT_SECRET, JWT_SECRET_KEY, DATA_DIR

4) Run the server
```bash
mcp-anywhere serve http

Alternatively, if you installed as a Python module:

python -m mcp_anywhere serve http

5. Access the gateway at http://localhost:8000 (default port) unless overridden by config

Production/deployment notes:

• For Fly.io deployment, follow the provided deployment guide in the project documentation.
• If using Docker, configure env values in a .env file and start via Docker Compose as described in the README.

Notes and tips:

• Environment variables control authentication (Google OAuth, JWT), credentials, and data paths. Keep SECRET_KEY and JWT keys secure.
• Secret files (JSON, PEM, certificate files, etc.) are stored encrypted at rest and mounted read-only into containers; environment variables are automatically populated with file paths.
• Tool discovery is automated from GitHub repositories; you can enable/disable tools per server and manage access via centralized credentials.
• Docker-based isolation helps ensure tools run in controlled environments; monitor container health and remount strategies as needed.
• If you’re using Claude Desktop integration, you’ll typically refer to the URL of the MCP Anywhere gateway and the CLI/uv configuration shown in the documentation.
• When running locally, you can use the CLI to connect to the gateway and interact with the server’s tools or run the STDIO server for Claude integration.
• Common issues: missing environment values, invalid OAuth credentials, or insufficient permissions for secret file storage. Review DATA_DIR and secret storage paths if you encounter file mounting errors.
• For troubleshooting, view server logs (docker logs or application logs) and ensure the gateway process has network access to required GitHub repos and external APIs.