A CLI host application that enables Large Language Models (LLMs) to interact with external tools through the Model Context Protocol (MCP). Currently supports Openai, Azure Openai, Deepseek and Ollama models.

English | 简体中文

• Interactive conversations with multipe LLM models
• Support for multiple concurrent MCP servers
• Dynamic tool discovery and integration
• Configurable MCP server locations and arguments
• Configurable message history window for context management
• Monitor/trace error from server side
• Support Sampling, Roots, Elicitation, retrievling Resource, Prompts
• Support runtime exclude specific tool
• Show MCP server card when connected

• [2025-07-18] Support Streamable HTTP mcp server, OAuth process not support yet
• [2025-07-02] Support Elicitation
• [2025-06-27] Deal with Prompts in MCP server: Link
• [2025-06-20] Deal with Resources in MCP server: Link

1. For Openai and Deepseek:

export OPENAI_API_KEY='your-api-key'

By default for Openai the base_url is "https://api.openai.com/v1" For deepseek it's "https://api.deepseek.com", you can change it by --base-url

2. For Ollama, need setup firstly:

• Install Ollama from https://ollama.ai
• Pull your desired model:

ollama pull mistral

• Ensure Ollama is running:

ollama serve

3. For Azure Openai:

export AZURE_OPENAI_DEPLOYMENT='your-azure-deployment'
export AZURE_OPENAI_API_KEY='your-azure-openai-api-key'
export AZURE_OPENAI_API_VERSION='your-azure-openai-api-version'
export AZURE_OPENAI_ENDPOINT='your-azure-openai-endpoint'

4. For Google Gemini

export GEMINI_API_KEY='your-gemini-api-token'

pip install mcp-cli-host

MCPCLIHost will automatically find configuration file at ~/.mcp.json. You can also specify a custom location using the --config flag:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/tmp/foo.db"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/tmp"
      ]
    }
  }
}

Each MCP server entry requires:

• command: The command to run (e.g., uvx, npx)
• args: Array of arguments for the command:
  • For SQLite server: mcp-server-sqlite with database path
  • For filesystem server: @modelcontextprotocol/server-filesystem with directory path

{
  "mcpServers": {
    "github": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {"Authorization": "Bearer <your PAT>"}
    }
  }
}

MCPCLIHost is a CLI tool that allows you to interact with various AI models through a unified interface. It supports various tools through MCP servers.

Models can be specified using the --model (-m) flag:

• Deepseek: deepseek:deepseek-chat
• OpenAI: openai:gpt-4
• Ollama models: ollama:modelname
• Azure Openai: azure:gpt-4-0613
• Gemini: gemini:gemini-2.5-flash

# Use Ollama with Qwen model
mcpclihost -m ollama:qwen2.5:3b

# Use Deepseek
mcpclihost -m deepseek:deepseek-chat --sys-prompt 'You are a slightly playful assistant, please answer questions in a cute tone!'

• --config string: Config file location (default is $HOME/mcp.json)
• --debug: Enable debug logging
• --message-window int: Number of messages to keep in context (default: 10)
• -m, --model string: Model to use (format: provider:model) (default "anthropic:claude-3-5-sonnet-latest")
• --base-url string: Base URL for OpenAI API (defaults to api.openai.com)
• --roots string: MCP clients to expose filesystem “roots” to servers
• --sys-prompt string: System prompt

While chatting, you can use:

• /help: Show available commands
• /tools: List all available tools
• /exclude_tool tool_name: Exclude specific tool from the conversation
• /resources: List all available resources
• /get_resource: Get specific resources by uri, example: /get_resource resource_uri
• /prompts: List all available prompts
• /get_prompt: Get specific prompt by name, example: /get_prompt prompt_name
• /servers: List configured MCP servers
• /history: Display conversation history
• /quit: Exit at any time

MCPCliHost can work with any MCP-compliant server. For examples and reference implementations, see the MCP Servers Repository.

• In scenario of Sampling and Elicitation, when typing "Ctrl+c", the process will crash with something like asyncio.exceptions.CancelledError, will be resolved later.

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.