MCPCLIHost

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.

- 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
- display image from tool call output

- [2025-11-26] Support display image from tool call output
- [2025-07-18] Support Streamable HTTP mcp server, OAuth process not support yet
- [2025-07-02] Support Elicitation
- [2025-06-27] Deal withPromptsin MCP server:Link
- [2025-06-20] Deal withResourcesin MCP server:Link

export OPENAI_API_KEY='your-api-key'

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

- Install Ollama fromhttps://ollama.ai
- Pull your desired model:

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'

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

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

- command: The command to run (e.g.,uvx,npx)
- args: Array of arguments for the command:

- For SQLite server:mcp-server-sqlitewith database path
- For filesystem server:@modelcontextprotocol/server-filesystemwith directory path

Remote mcp server(only support Streamable HTTP)

{ "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

- /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 theMCP Servers Repository.

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

Instead of direct calling MCP tools, mcpcode server transforms MCP tool calls into TypeScript programs, enabling smarter, lower-latency orchestration by LLMs.