inkeep

Inkeep MCP serves as a universal bridge that lets your AI agent read documentation content protected behind Inkeep-powered widgets. Start the MCP server using the Python entry point shown in the repository, then configure your agent (e.g., Gemini CLI or Claude Desktop) to load this MCP server as a local provider. The server simulates a browser session to locate Inkeep configuration on target sites and streams answers back to your agent, enabling question answering and interactive chat against docs from Langfuse, Render, Clerk, Neon, and more. You can use the CLI to ask questions directly or wire the server into your agent’s environment so it can query docs on demand.

To use with agents, add the inkeep MCP entry to your agent’s MCP config (e.g., Gemini or Claude). The config should specify the Python command, the path to mcp_server.py, and optional environment flags like PYTHONUNBUFFERED for clean streaming. Once configured, your agent can send queries to the server, which will fetch, process, and return the results from the supported documentation sites.

Prerequisites:

• Python 3.8+ installed on your system
• Git
• Internet access to install dependencies

Install steps:

1. Clone the repository git clone https://github.com/xu-xiang/inkeep-mcp.git cd inkeep-mcp

2. Install Python dependencies pip install -r requirements.txt

3. Run the MCP server (example path; adjust to your environment) python3 mcp_server.py

4. (Optional) Integrate with your agent by updating its MCP config to point at the server as shown in the repository examples.

Notes and tips:

• The MCP config uses a Python-based server; ensure you provide the absolute path to mcp_server.py in your agent config.
• The environment variable PYTHONUNBUFFERED=1 helps ensure real-time streaming of responses.
• If you encounter path issues, verify that the path in args is correct and accessible by the agent process.
• This MCP relies on simulating a browser session to obtain Inkeep configuration from target sites; some sites may require additional handling or rate limiting in your environment.
• When debugging, run the server in a console to observe logs and stdout/stderr for troubleshooting.