MCP Host CLI
@VyacheslavVanin
Local http server that proxies requests o LLMs and uses mcp-servers if needed
Overview
What is MCP Host CLI?
It is a FastAPI server for chat interactions with LLMs, using OpenAI-compatible providers. It is configured via environment variables and CLI arguments, and supports tool approval workflows.
How to use MCP Host CLI?
Configure via environment variables (e.g., LLM_API_KEY, LLM_MODEL) or CLI arguments (e.g., --model, --port). Start the server using uv run main.py with desired options. Send requests to endpoints like POST /user_request and POST /approve.
Key features of MCP Host CLI
- FastAPI server for chat interactions with LLMs
- Configurable via environment variables and CLI arguments
- Supports OpenAI-compatible providers (e.g., OpenRouter)
- Tool approval workflow with request/response endpoints
- Streaming mode and retry on LLM errors
Use cases of MCP Host CLI
- Building a backend for chat interfaces with LLMs
- Integrating LLM chat with tool approval mechanisms
- Running a local or remote LLM chat server with OpenAI-compatible API
FAQ from MCP Host CLI
How do I configure the LLM model?
Use the LLM_MODEL environment variable or the --model CLI argument. Default is qwen2.5-coder:latest.
How do I provide the API key?
Set the LLM_API_KEY environment variable or use the --api-key-file CLI argument pointing to a file with the key.
What LLM providers are supported?
Only openai provider is supported, but the base URL can be changed (default OpenRouter).
How do I approve a tool call?
Send a POST request to /approve with request_id and approve boolean in the JSON body.
Where are debug messages logged?
If --debug-messages is enabled, logs are written to /tmp/llm-requester.messages.log.
