GitHub - Flux159/mcp-chat: Open Source Generic MCP Client for testing & evaluating mcp servers and agents (original) (raw)
mcp-chat
Open Source Generic MCP Client for testing & evaluating mcp servers and agents
Quickstart
Make sure that you have ANTHROPIC_API_KEY exported in your environment or in a .env file in the root of the project. You can get an API key by signing up at the Anthropic Console keys page.
Simple use case that spawns an interactive chat prompt with the filesystem MCP server from CLI:
npx mcp-chat --server "npx -y @modelcontextprotocol/server-filesystem /Users/$USER/Desktop"
This will open up a chat prompt that you can use to interact with the servers and chat with an LLM.
Config
You can also just specify your claude_desktop_config.json (Mac):
npx mcp-chat --config "~/Library/Application Support/Claude/claude_desktop_config.json"
Or (Windows):
npx mcp-chat --config "%APPDATA%\Claude\claude_desktop_config.json"
Web mode
MCPChatWebDemo.mov
You can also run mcp-chat in web mode by specifying the --web flag (make sure to have ANTHROPIC_API_KEY exported in your environment):
In web mode, you can start new chats, send messages to the model, and dynamically configure the mcp servers via the UI - no need to specify on the command line. In addition, chats created via the Web UI are saved to ~/.mcpchats/chats just like chats created via the CLI.
Features
- Run via CLI in interactive mode or directly pass prompts with
-p - Web mode to chat with models via a web interface
--web - Connect to any MCP server (JS, Python, Docker) in production or during development
- Choose between models with
-m - Customize system prompt with
--system - Saves chat history with settings in
~/.mcpchat/chatsincluding web chats - Save and restore commands in
~/.mcpchat/history - View tool call output and arguments directly in chat to help debug mcp servers
CLI Usage
Run prompts via CLI with the -p flag:
npx mcp-chat --server "npx mcp-server-kubernetes" -p "List the pods in the default namespace"
This runs the prompt with the kubenertes mcp-server & exits after the response is received on stdout.
Choose a model to chat with via CLI with the -m flag:
npx mcp-chat --server "npx mcp-server-kubernetes" -m "claude-3.5"
Uses the model claude-3.5 to chat with. Note that currently only Anthropic models are supported.
Custom system prompt:
--system flag can be used to specify a system prompt:
npx mcp-chat --system "Explain the output to the user in pirate speak." --server "npx mcp-server-kubernetes" -p "List the pods in the default namespace"
ENV vars
Mcp-chat supports passing through env vars to mcp servers. This is not standardized in the mcp spec though and other clients may not do this with dynamic env var values - read more here about standard stdio mcp clients.
KUBECONFIG="~/.kube/config" npx mcp-chat --server "npx mcp-server-kubernetes"
For developers of mcp-servers
You can pass in a local build of a python or node mcp-server to test it out with mcp-chat:
Node JS:
Directly executing built script
npx mcp-chat --server "/path/to/mcp-server-kubernetes/dist/index.js"
Using node / bun
npx mcp-chat --server "node /path/to/mcp-server-kubernetes/dist/index.js"
Python:
Python: Using uv
npx mcp-chat --server "uv --directory /path/to/mcp-server-weather/ run weather.py"
Using python / python3 - make sure to run in venv or install deps globally
npx mcp-chat --server "/path/to/mcp-server-weather/weather.py"
Development
Install dependencies & run the CLI:
git clone https://github.com/Flux159/mcp-chat bun install bun run dev
To develop mcp-chat while connecting to an mcp-server, make a build & run the CLI with the server flag:
npm run build && node dist/index.js --server "npx mcp-server-kubernetes" -p "List the pods in the default namespace"
Testing:
Building:
Publishing:
Publishing Docker:
Project Structure
├── src/
│ ├── index.ts # Main client implementation & CLI params
│ ├── constants.ts # Default constants
│ ├── interactive.ts # Interactive chat prompt handling & logic
├── test/ # Test files
│ ├── cli.test.ts # Test CLI params
│ ├── config.test.ts # Test config file parsing
Publishing new release
Go to the releases page, click on "Draft New Release", click "Choose a tag" and create a new tag by typing out a new version number using "v{major}.{minor}.{patch}" semver format. Then, write a release title "Release v{major}.{minor}.{patch}" and description / changelog if necessary and click "Publish Release".
This will create a new tag which will trigger a new release build via the cd.yml workflow. Once successful, the new release will be published to npm. Note that there is no need to update the package.json version manually, as the workflow will automatically update the version number in the package.json file & push a commit to main.