概览
What is mcp-chat?
mcp-chat is an open source generic MCP client for testing and evaluating MCP servers and agents. It provides a CLI and web interface for interacting with MCP servers through an LLM.
How to use mcp-chat?
Run via npx mcp-chat with flags like --server, --config, --web, -m (model), -p (prompt), or --system (system prompt). Requires the ANTHROPIC_API_KEY environment variable set. For web mode, use --web and configure servers through the UI.
Key features of mcp-chat
- Run CLI in interactive mode or with
-pflag - Web mode for chatting via a web interface
- Connect to any MCP server (JS, Python, Docker)
- Choose between Anthropic models with
-m - Customize system prompt with
--system - Saves chat history to
~/.mcpchat/chats
Use cases of mcp-chat
- Test MCP servers during development (local builds)
- Run one‑off prompts with MCP servers (e.g., Kubernetes)
- Interactive chat with an LLM and connected tools
- Debug tool call output and arguments
FAQ from mcp-chat
What LLM models are supported?
Currently only Anthropic models (e.g., claude-3.5) are supported.
How do I connect to a local MCP server?
Use --server with a command (e.g., npx mcp-server-kubernetes) or pass a local script (e.g., node /path/to/index.js or uv run weather.py).
Where is chat history saved?
Chats are saved in ~/.mcpchat/chats (both CLI and web mode).
Can I run mcp-chat without a CLI server argument?
Yes, in web mode (--web) you can dynamically configure MCP servers via the UI.
Do I need an API key?
Yes, set the ANTHROPIC_API_KEY environment variable or place it in a .env file.
