Happy slice of mcp pizzaMCP.pizza

mcp-chat

MCP.Pizza Chef: Flux159

A chat program you run yourself, in a terminal or a browser tab, pointed at whichever MCP servers you want to exercise. It prints each tool call and its result, so you can see exactly what a server did. Only Claude models are supported, and a paid Anthropic account key is demanded at startup in every mode. Two advertised switches do nothing: agent mode is never read, and evaluation mode prints one line and stops.

Coding

Use This MCP client To

Test a new MCP server before adding it to my app See exactly which tool ran and what it returned Reuse my Claude Desktop server list in a terminal chat Run a single question against a server and get the answer Pick an earlier conversation back up where I left off Chat with my servers from a browser tab instead

README

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"

mcp-chat FAQ

Do I need an account or key?
Yes. A paid Anthropic account key is required and is checked the moment it starts, including in browser mode. There is no free tier and no bring-your-own-model option.
Can I use it with ChatGPT or Gemini models?
No. Only Claude models work. The default is a Claude 3.7 Sonnet build, and the shorthand model name shown in the project's own example is not a real one.
Can I use this to try out a server I just downloaded?
Yes. Point it at the command that starts the server and you get a chat window with that server's tools available straight away.
Is the browser mode safe to leave running?
Be careful. It has no password or login of any kind and listens on every network connection your machine has, so anyone on the same network can open it and drive your connected servers.
Does it ask before running a tool?
No. There is no confirmation step, so a connected server that deletes files or spends money will be allowed to do so without asking you first.
Does the evaluation mode work?
No. The switch is accepted and prints a message, but the feature is an unfinished placeholder in the code. The agent-mode switch is accepted and never used at all.
Where do my conversations get stored?
In a hidden folder in your home directory, as plain readable files, written after every response.
How hard is setup?
Fine if you are comfortable with a terminal: one command plus your account key. It can also read the server list you already set up for Claude Desktop.