MCP CLI client

What is MCP CLI client?

MCP CLI client is a command-line program that runs LLM prompts and implements the Model Context Protocol (MCP) client. It works on any platform that supports Python and pip, and is designed for developers who want to interact with MCP‑compatible servers directly from their terminal, as an alternative to Claude Desktop.

How to use MCP CLI client?

Install via pip install mcp-client-cli, then create a ~/.llm/config.json file that specifies your LLM provider (e.g., OpenAI, Groq, or a local llama.cpp model) and one or more MCP servers. Run the CLI with llm "your prompt". You can pipe input from files or commands, use image input, apply prompt templates, and continue previous conversations with the c prefix.

Key features of MCP CLI client

• Run prompts with any LLM provider (OpenAI, Groq, local llama.cpp)
• Connect to any MCP‑compatible server via a config file
• Tool confirmation prompts and a --no-confirmations flag
• Image input via piping image files or clipboard
• Prompt templates for common tasks (review, commit, YouTube)
• Clipboard support across Windows, macOS, and Linux

Use cases of MCP CLI client

• Querying the web or performing searches via MCP tools (e.g., Brave Search)
• Analyzing images with multimodal LLMs by piping image files
• Automating code reviews or commit message generation with prompt templates
• Building scriptable workflows that combine LLM reasoning with MCP tool execution
• Continuing a prior conversation using the c prefix

FAQ from MCP CLI client

What does MCP CLI client do differently from Claude Desktop?

MCP CLI client is a terminal‑based MCP client that works with any LLM provider, whereas Claude Desktop is a graphical application tied to Anthropic’s models. The CLI gives you flexibility to choose your own model and automate tasks in scripts.

Which LLM providers and models are supported?

The configuration supports any OpenAI‑compatible provider (set provider and base_url), Groq, and local models via llama.cpp. You can override the model at runtime with --model.

How do I add an MCP server to MCP CLI client?

Add an entry under mcpServers in ~/.llm/config.json specifying the command, arguments, environment variables, and optional settings like requires_confirmation or exclude_tools.

Can I use MCP CLI client without paying?

The tool itself is free and open‑source (installed via pip). However, you will need API keys or local model resources for the LLM and any MCP servers you connect to; those may have their own costs.

Are there any known limitations?

The README does not list specific limitations. Tool confirmation is required by default for certain tools, but you can bypass it with --no-confirmations. Intermediate messages can be suppressed with --no-intermediates for scripting.