MCP Link provides a Python-based MCP SSE backend that securely executes local or remote tools for AI agents. The server acts as the runtime that receives tool calls from compliant AI clients, validates permissions, and runs the requested actions in a controlled environment (including optional Docker sandboxing). The browser extension and recipes fetchable from the remote manifest define which tools are available and how to call them, while the server executes those calls and returns structured responses. You can use the extension to discover tools on supported sites, and when you enable local tools, the MCP Link server is the component that actually runs them in your environment. The system is designed to give AI models plugin-like access to your tools without exposing your entire system to the cloud.

Prerequisites:

• Python 3.9+ installed on the host
• Git to clone the repository
• Optional: Docker if you want sandboxed tool environments

Step-by-step installation:

1. Clone the repository git clone https://github.com/AuraFriday/mcp-link.git cd mcp-link

2. Set up a Python virtual environment python3 -m venv venv source venv/bin/activate # on macOS/Linux

   Windows: venv\Scripts\activate

3. Install server dependencies (venv) pip install -r server/requirements.txt

4. Prepare the server configuration (optional)

   • Create a config.json as needed, or rely on defaults. Example placeholder: { "port": 8000, "tools": {} }

5. Run the MCP SSE backend server (venv) python -m server

   Or specify a config path if your module supports it, e.g.:

   (venv) MCP_CONFIG=config.json python -m server

6. (Optional) Run with Docker for sandboxed environments docker run -i --rm mcp-link-server:latest

Notes:

• Ensure any local tools you want to expose are installed on the host or accessible to the server process.
• If you modify configuration, restart the server to apply changes.

Environment variables and configuration tips:

• MCP_SERVER_CONFIG: Path or inline JSON configuration for allowed tools and policies.
• LOG_LEVEL: Set to DEBUG during troubleshooting, INFO for normal operation.
• If using Docker sandboxing, ensure the host allows network access as needed and that containers have access to required binaries/tools.
• Tools are discovered via the remote manifest at https://cdn.aurafriday.com/config.json; ensure the MCP Link extension is configured to fetch and expose the desired tools.
• For enterprise setups, consider enabling per-tool usage approvals and logging as described in the MCP Link documentation.
• If you encounter permission errors, verify file permissions and that the Python process has access to the tool executables.
• When no UI is required, you can operate the server headlessly and control tool access exclusively through the extension and config.