LLM MCP Server

A Model Context Protocol (MCP) server that provides tools to query Large Language Models (LLMs) using the llm-querier library. Currently supports OpenAI and Google Gemini.

Installation

Option A: Global Installation

npm install -g @edjl/llm-mcp

Option B: Use with npx (no installation required)

npx -y @edjl/llm-mcp

Configuration

Set the following environment variables:

OpenAI Configuration

• OPENAI_API_KEY: Your OpenAI API key (required for OpenAI tool)
• OPENAI_MODEL: OpenAI model to use (default: o3)

Google Gemini Configuration

• GEMINI_API_KEY: Your Google Gemini API key (required for Gemini tool)
• GEMINI_MODEL: Gemini model to use (default: gemini-2.5-pro)

Note: You can configure just one provider or both. The server will only enable tools for configured providers.

Available Tools

llm_ask_openai

Ask a single query prompt to OpenAI. Provide as much context as possible. This is a single call - no conversation state is maintained.

Parameters:

• prompt (required): The query prompt to send to OpenAI
• context (optional): Array of additional context strings to enhance the prompt
• examples (optional): Array of examples to guide the response
• images (optional): Array of image URLs or base64 encoded images
• scrapeUrls (optional): Array of URLs to scrape and include as context
• fileUrls (optional): Array of file URLs to download and include as context

llm_ask_gemini

Ask a single query prompt to Google Gemini. Provide as much context as possible. This is a single call - no conversation state is maintained.

Parameters:

• prompt (required): The query prompt to send to Google Gemini
• context (optional): Array of additional context strings to enhance the prompt
• examples (optional): Array of examples to guide the response
• images (optional): Array of image URLs or base64 encoded images
• videos (optional): Array of video URLs (Google AI supports video input)
• scrapeUrls (optional): Array of URLs to scrape and include as context
• fileUrls (optional): Array of file URLs to download and include as context

Usage with Cursor

Add to your Cursor settings:

Option A: Global Installation

{
  "mcpServers": {
    "llm-mcp": {
      "command": "llm-mcp",
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key",
        "OPENAI_MODEL": "o3",
        "GEMINI_API_KEY": "your-gemini-api-key",
        "GEMINI_MODEL": "gemini-2.5-pro"
      }
    }
  }
}

Option B: Using npx

{
  "mcpServers": {
    "llm-mcp": {
      "command": "npx",
      "args": ["-y", "@edjl/llm-mcp"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key",
        "OPENAI_MODEL": "o3",
        "GEMINI_API_KEY": "your-gemini-api-key",
        "GEMINI_MODEL": "gemini-2.5-pro"
      }
    }
  }
}

Examples

Basic Query to OpenAI

const result = await use_mcp_tool({
  server_name: "llm-mcp",
  tool_name: "llm_ask_openai",
  arguments: {
    prompt: "Explain the concept of quantum computing in simple terms"
  }
});

Query with Context and Examples

const result = await use_mcp_tool({
  server_name: "llm-mcp",
  tool_name: "llm_ask_gemini",
  arguments: {
    prompt: "Write a haiku about programming",
    context: ["Focus on the debugging process", "Make it humorous"],
    examples: ["Bugs hide in the code / Like ninjas in the shadows / Coffee is my sword"]
  }
});

Query with Web Scraping

const result = await use_mcp_tool({
  server_name: "llm-mcp",
  tool_name: "llm_ask_openai",
  arguments: {
    prompt: "Summarize the main points from this article",
    scrapeUrls: ["https://example.com/article"]
  }
});

Query with Images (Vision Models)

const result = await use_mcp_tool({
  server_name: "llm-mcp",
  tool_name: "llm_ask_gemini",
  arguments: {
    prompt: "What's in this image?",
    images: ["https://example.com/image.jpg"]
  }
});

Notes

• This MCP server uses the llm-querier library under the hood
• Each query is independent - no conversation history is maintained
• The server only loads tools for providers that have API keys configured
• For more advanced usage, refer to the llm-querier documentation

License

MIT