MCP LLM

An MCP server that provides access to LLMs using the LlamaIndexTS library.

Features

This MCP server provides the following tools:

• generate_code: Generate code based on a description
• generate_code_to_file: Generate code and write it directly to a file at a specific line number
• generate_documentation: Generate documentation for code
• ask_question: Ask a question to the LLM

Installation

Update your MCP config to add the mcp-llm server:

{
  "mcpServers": {
    "llm": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-llm"
      ],
      "env": {
        "LLM_MODEL_NAME": "deepseek-r1:7b-qwen-distill-q6_k_l",
        "LLM_MODEL_PROVIDER": "ollama",
        "LLM_BASE_URL": "http://localhost:11434",
        "LLM_ALLOW_FILE_WRITE": "true",
        "LLM_TIMEOUT_S": "240"
      },
      "disabled": false,
      "autoApprove": [
        "generate_code",
        "generate_documentation",
        "ask_question",
        "generate_code_to_file"
      ],
      "timeout": 300
    },
  }
}

Available Scripts

• npm run build - Build the project
• npm run watch - Watch for changes and rebuild
• npm start - Start the MCP server
• npm run example - Run the example script
• npm run inspector - Run the MCP inspector

Configuration

The MCP server is configurable using environment variables:

Required Environment Variables

• LLM_MODEL_NAME: The name of the model to use (e.g., qwen2-32b:q6_k, anthropic.claude-3-7-sonnet-20250219-v1:0)
• LLM_MODEL_PROVIDER: The model provider (e.g., bedrock, ollama, openai, openai-compatible)

Optional Environment Variables

• LLM_BASE_URL: Base URL for the model provider (e.g., https://ollama.internal, http://my-openai-compatible-server.com:3000/v1)
• LLM_TEMPERATURE: Temperature parameter for the model (e.g., 0.2)
• LLM_NUM_CTX: Context window size (e.g., 16384)
• LLM_TOP_P: Top-p parameter for the model (e.g., 0.85)
• LLM_TOP_K: Top-k parameter for the model (e.g., 40)
• LLM_MIN_P: Min-p parameter for the model (e.g., 0.05)
• LLM_REPETITION_PENALTY: Repetition penalty parameter for the model (e.g., 1.05)
• LLM_SYSTEM_PROMPT_GENERATE_CODE: System prompt for the generate_code tool
• LLM_SYSTEM_PROMPT_GENERATE_DOCUMENTATION: System prompt for the generate_documentation tool
• LLM_SYSTEM_PROMPT_ASK_QUESTION: System prompt for the ask_question tool
• LLM_TIMEOUT_S: Timeout in seconds for LLM requests (e.g., 240 for 4 minutes)
• LLM_ALLOW_FILE_WRITE: Set to true to allow the generate_code_to_file tool to write to files (default: false)
• OPENAI_API_KEY: API key for OpenAI (required when using OpenAI provider)

Manual Install From Source

1. Clone the repository
2. Install dependencies:

npm install

3. Build the project:

npm run build

4. Update your MCP configuration

Using the Example Script

The repository includes an example script that demonstrates how to use the MCP server programmatically:

node examples/use-mcp-server.js

This script starts the MCP server and sends requests to it using curl commands.

Examples

Generate Code

{
  "description": "Create a function that calculates the factorial of a number",
  "language": "JavaScript"
}

Generate Code to File

{
  "description": "Create a function that calculates the factorial of a number",
  "language": "JavaScript",
  "filePath": "/path/to/factorial.js",
  "lineNumber": 10,
  "replaceLines": 0
}

The generate_code_to_file tool supports both relative and absolute file paths. If a relative path is provided, it will be resolved relative to the current working directory of the MCP server.

Generate Documentation

{
  "code": "function factorial(n) {\n  if (n <= 1) return 1;\n  return n * factorial(n - 1);\n}",
  "language": "JavaScript",
  "format": "JSDoc"
}

Ask Question

{
  "question": "What is the difference between var, let, and const in JavaScript?",
  "context": "I'm a beginner learning JavaScript and confused about variable declarations."
}

License

• MIT LICENSE