mcp-chat CLI
A terminal and browser client opens a chat between an Anthropic model and any MCP server so its tools can be tested.
- 139 ★ on the repository
- Last commit
Extended access
Why this level
- A connected server gets access to files and the network, and prompts go to the Anthropic API.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/mcp-chatDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- A connected server gets access to files and the network, and prompts go to the Anthropic API.
This is third-party code. Review the repository files before installing.
What it does
mcp-chat starts an MCP server, connects it to an Anthropic model and opens an interactive chat. Replies show tool calls with their arguments and results, so the dialogue makes it clear what the server actually does. The server is set with `--server` or with a configuration file in the `claude_desktop_config.json` format. Local servers in JS, Python and Docker work, as do remote ones over Streamable HTTP. The `-p` flag runs one prompt and exits, `--web` opens a web interface. Chat history is stored in `~/.mcpchat/chats`.
Who it is for. MCP server authors, and anyone evaluating a third-party server who wants to see its tools in action before wiring it into a working agent.
Good fit when
- A fresh MCP server needs a run on real prompts with tool calls visible.
- A Claude Desktop configuration should be checked apart from the app itself.
- A script runs a single request against a server with the `-p` flag.
- A remote HTTP server is tested without installing a graphical client.
Not a fit when
- A non-Anthropic model is required, since only its models are supported.
- An automated regression suite for the server is needed, not a manual chat.
- No Anthropic key is at hand.
Example request
Connect the Kubernetes MCP server, list the pods in the default namespace and show the tools that were called.Limitations
An `ANTHROPIC_API_KEY` is required and every model request is billed. Servers launched from a command get access to your file system and network, so connect only vetted ones. The README example exposes the whole desktop; pass a narrow folder instead. Only Anthropic models are supported and another provider cannot be chosen. Chats are saved to disk together with settings and may contain tool results with working data, so clear the history directory periodically.
How to disable. Close the chat session and delete the saved history in `~/.mcpchat` if it is not needed. The client creates no permanent settings in the agent.
FAQ
How do I connect several servers?
Pass a Claude Desktop format configuration file through `--config`. README shows paths for Mac and Windows.
How do I run one prompt without a chat?
Add `-p "prompt text"`. The client prints the answer to stdout and exits.
Is there a web interface?
Yes, `npx mcp-chat --web`. Servers are configured in a form and chats are saved the same way as from the console.
Does a remote server work?
Yes. Pass a Streamable HTTP URL to `--server` instead of a command.
Where is history stored?
Chats are in `~/.mcpchat/chats` and the command history in `~/.mcpchat/history`. Web chats are saved there too.
Discussion
No questions or reviews yet.
Related
CLI toolsA command-line debugger for MCP servers from Giant Swarm: a REPL for tools, OAuth 2.1 and an MCP server mode for Cursor
A transparent proxy between an AI client and an MCP server: shows every JSON-RPC frame in the terminal, checks a capture in CI and exports spans
An async Python MCP client with a CLI and web UI that connects Claude to servers over stdio and SSE.
The openmcp CLI and a VS Code extension for debugging MCP servers: tools, prompts and resources, chat with a model, test cases
A lightweight CLI to discover, inspect and call MCP server tools straight from the shell
A teaching MCP server in Python: Hacker News search, GitHub repository details and examples of model sampling requests
README in short
The quickstart shows a launch with the filesystem server. Later sections cover configuration, web mode, the `-p`, `-m` and `--system` flags and remote servers.