fastchat

Fastchat MCP provides a Python-based client that connects to MCP servers using multiple transfer protocols (primarily stdio and HTTPStream). It is designed to work with integrated language models (notably OpenAI-compatible models) and offers a modular interface to configure and communicate with MCP servers. The project exposes a command-line client via the mcp[cli] integration, enabling you to interact with configured MCP servers, send prompts, and stream responses through standard protocols. You can leverage its configuration to route conversations through a chosen server, manage model selection, and handle streaming or standard I/O interactions as supported by the backend server.

Prerequisites:

• Python 3.11 or newer
• Access to install Python packages (pip)
• OpenAI API key (if using OpenAI models)

Installation steps:

1. Create and activate a Python environment (optional but recommended):

python -m venv venv
source venv/bin/activate  # on Unix/macOS
venv\Scripts\activate     # on Windows

2. Install the MCP client package:

pip install fastchat-mcp

3. Prepare required environment variables (example):

export OPENAI_API_KEY=your-openai-key
export CRIPTOGRAFY_KEY=your-cryptography-key

4. Run the MCP server client (as configured in mcp_config):

python -m fastchat_mcp

(Adjust as needed if you use a different entry point or environment setup.)

Tips and notes:

• Ensure Python 3.11+ is used to meet the dependencies of fastchat-mcp.
• Set OPENAI_API_KEY in your environment if you plan to access OpenAI models; the project notes indicate OpenAI provider support (default provider).
• The CRIPTOGRAFY_KEY is used for token data storage and related cryptographic operations; keep it secure and do not expose in logs.
• The MCP configuration file fastchat.config.json should be placed at the repository root and define your available MCP servers (protocols: httpstream or stdio).
• The project currently emphasizes stdio and HTTPStream protocols; SSE is not supported.
• If you encounter issues, verify that the environment variables are loaded in the same process that runs the MCP client and that network access to OpenAI endpoints is allowed.
• For debugging, run with verbose logging if available in the CLI to inspect MCP server connection details and protocol handshakes.