📌Overview

Purpose:
Provide a simple command-line interface (CLI) for interacting with Large Language Models (LLMs) using the Model Context Protocol (MCP), supporting multiple LLM providers and MCP-compatible tool servers from the terminal.

Overview:
MCP CLI client is a lightweight CLI tool that enables running LLM prompts and integrating with any MCP-compatible servers, allowing users to access advanced AI capabilities from their terminal. It serves as an alternative to graphical clients, supporting a diverse range of LLM providers (including OpenAI, Groq, and local models) and can be configured flexibly via a user-friendly JSON file.

Key Features:

• Multi-provider LLM support:
  Seamlessly connects to various LLM providers (OpenAI, Groq, local llama.cpp, etc.) through unified configuration, allowing users to work with the model of their choice.

• MCP protocol integration:
  Works with any MCP-compatible server to trigger external tools (e.g., web search, YouTube summarization), enhancing AI responses with real-time data and automation from the terminal.

• Flexible and extensible configuration:
  Supports JSON-based config files with environment variable overrides, customizable prompts, and tool management, enabling users to tailor the CLI to their workflows and security preferences.

• Rich input and prompt features:
  Allows piped input (including images), predefined prompt templates, clipboard support (for both text and image), and conversation continuation, making the CLI versatile for a range of tasks and scripting scenarios.

• Convenient output and scripting options:
  Offers flags for controlling output format, tool usage, confirmation requirements, and integration in shell pipelines or automation scripts.

---

MCP CLI Client

A simple command-line program for running LLM prompts and acting as a Model Context Protocol (MCP) client. You can use any MCP-compatible servers conveniently from your terminal. This tool provides an alternative to other clients, such as Claude Desktop, and supports multiple LLM providers like OpenAI, Groq, or a local LLM model via llama.cpp.

---

Installation and Configuration

1. Install

pip install mcp-client-cli

2. Configure

Create a ~/.llm/config.json file to set up your LLM and MCP servers. Example:

{
  "systemPrompt": "You are an AI assistant helping a software engineer...",
  "llm": {
    "provider": "openai",
    "model": "gpt-4",
    "api_key": "your-openai-api-key",
    "temperature": 0.7,
    "base_url": "https://api.openai.com/v1"
  },
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"],
      "requires_confirmation": ["fetch"],
      "enabled": true,
      "exclude_tools": []
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": {
        "BRAVE_API_KEY": "your-brave-api-key"
      },
      "requires_confirmation": ["brave_web_search"]
    },
    "youtube": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/adhikasp/mcp-youtube", "mcp-youtube"]
    }
  }
}

• Use requires_confirmation for tools needing user approval before execution.
• LLM API keys can be set via the config or the environment variable LLM_API_KEY or OPENAI_API_KEY.
• Config file can be in ~/.llm/config.json or $PWD/.llm/config.json.
• Comments are supported in the config file using //.

3. Run

llm "What is the capital city of North Sumatra?"

---

Usage

Basic Queries

llm What is the capital city of North Sumatra?

You can omit quotes unless special shell characters are involved.

Piping from files or commands is supported:

echo "What is the capital city of North Sumatra?" | llm
echo "Given a location, tell me its capital city." > instructions.txt
cat instruction.txt | llm "West Java"

Image Input

Analyze images by piping them:

cat image.jpg | llm "What do you see in this image?"
cat screenshot.png | llm "Is there any error in this screenshot?"

Prompt Templates

Use prompt templates with a p prefix:

llm --list-prompts                   # List prompt templates
llm p review                         # Review git changes
llm p commit                         # Generate commit message
llm p yt url=https://youtube.com/... # Summarize a YouTube video

Using Tools

Example showing a tool-triggering interaction:

llm What is the top article on hackernews today?

You may be prompted to confirm actions:

Confirm tool call? [y/n]: y

To bypass confirmations, use:

llm --no-confirmations "What is the top article on hackernews today?"

To output only the final message in scripts:

llm --no-intermediates "What is the time in Tokyo right now?"

Conversation Continuation

Continue previous conversations with the c prefix:

llm asldkfjasdfkl
llm c what did i say previously?

Clipboard Support

Process clipboard content using the cb command:

• For text:

  llm cb
  llm cb "What language is this code written in?"
  

• For images:

  llm cb "What do you see in this image?"
  

• Combine with continuation:

  llm cb c "Tell me more about what you see"
  

Clipboard feature compatibility:

• Windows: Uses PowerShell.
• macOS: Uses pbpaste (text), pngpaste (for images, optional, install with brew install pngpaste).
• Linux: Requires xclip (sudo apt install xclip).

---

Additional Options

llm --list-tools           # List all tools
llm --list-prompts         # List prompt templates
llm --no-tools             # Run without tools
llm --force-refresh        # Refresh tool capabilities
llm --text-only            # Output raw text only
llm --show-memories        # Show user memories
llm --model gpt-4          # Override default model

---

Contributing

Contributions and bug reports are welcome!