Ultimate MCP Client CLI
An async Python MCP client with a CLI and web UI that connects Claude to servers over stdio and SSE.
- 171 ★ on the repository
- Last commit
Broad access
Why this level
- Connected MCP servers can run commands and read files with user rights.
- Port scanning and mDNS probe the local network for servers.
Install
Manual install
git clone https://github.com/Dicklesworthstone/ultimate_mcp_clientThen `uv venv --python 3.13` and `uv sync --all-extras` inside the folder. Run it with `mcpclient run --webui`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Connected MCP servers can run commands and read files with user rights.
- Port scanning and mDNS probe the local network for servers.
This is third-party code. Review the repository files before installing.
What it does
Ultimate MCP Client runs a conversation with Claude and routes tool calls to connected MCP servers. It supports the `stdio`, `sse` and streaming-http transports and discovers servers in local folders, over mDNS, by port scanning, in registries and in the Claude Desktop configuration. Conversations branch as a graph saved to JSON, and long history can be compressed with `/optimize`. It adds a tool-result cache, OpenTelemetry metrics and a live dashboard at `mcpclient run --dashboard`.
Who it is for. MCP server developers who need a debugging client with its own interface, logs and metrics.
Good fit when
- A new MCP server must be tested by hand through chat and a direct `/tool` call.
- You want to find every MCP server on the local network or on open ports.
- A tool-using conversation should branch so two paths can be compared.
- A Claude Desktop configuration needs to move into a standalone client.
Not a fit when
- You need a client for non-Anthropic models; the integration is built around Claude.
- A server is being attached to Claude Code or Cursor, which already have their own MCP client.
- Local port scanning is banned by network policy.
Example request
Connect the server from `.mcpclient_config/servers`, list its tools and call one of them with test parameters.Limitations
It needs Python 3.13 or newer and your own `ANTHROPIC_API_KEY`; Anthropic bills the requests. The license is MIT with an additional OpenAI and Anthropic rider whose text you should read in `LICENSE`. The `--host 0.0.0.0` flag exposes the web UI to the whole network, and connected servers can run commands. Configuration and conversations live in the project folder `.mcpclient_config/`, so keys and history sit next to the code and should not be committed. Python 3.13 may be missing on the system, and `uv` is installed separately.
How to disable. Stop the client and delete the cloned folder. Saved conversations and settings live in `.mcpclient_config/` in the project root.
FAQ
How do I start the web UI?
Run `mcpclient run --webui`. The `--host` and `--port` flags change the address and port.
How do I set the API key?
Export `ANTHROPIC_API_KEY` or put it in `.env` in the project root. The environment variable overrides the value in the configuration.
Which servers does the client find by itself?
Scripts in configured folders, local-network servers over mDNS, servers on ports in a set range and entries from remote registries.
How do I save a conversation?
`mcpclient export` writes a branch to JSON and `mcpclient import-conv` loads it back. Conversations are also saved automatically in the configuration folder.
Does the client work with non-Anthropic models?
The README describes integration with Claude through the official `anthropic` package. It names no other providers.
Discussion
No questions or reviews yet.
Related
CLI toolsAn MCP server that gives Claude Desktop and other clients file tools, a shell and sub-agents in the style of Claude Code
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 MCP server that gives an agent search over a large Claude Code guide, official docs, releases and ready-made templates
A terminal and browser client opens a chat between an Anthropic model and any MCP server so its tools can be tested.
A command-line debugger for MCP servers from Giant Swarm: a REPL for tools, OAuth 2.1 and an MCP server mode for Cursor
A lightweight CLI to discover, inspect and call MCP server tools straight from the shell
README in short
The client is written in Python with FastAPI for the web part and Typer for the CLI. The README details transports, server discovery, conversation management and observability.