OpenAI Complete MCP Server

An MCP (Model Context Protocol) server that provides a clean interface for LLMs to use text completion capabilities through the MCP protocol. This server acts as a bridge between an LLM client and any OpenAI's compatible API. The primary use case is for base models, as the server does not provide support for chat completions.

Features

Provides a single tool named "complete" for generating text completions
Properly handles asynchronous processing to avoid blocking
Implements timeout handling with graceful fallbacks
Supports cancellation of ongoing requests

Installation

# Clone the repository
git clone <repository-url>
cd mcp-openai-complete

# Install dependencies
pnpm install

# Build the project
pnpm run build

Configuration

The following environment variables are required:

OPENAI_API_KEY=your-hyperbolic-api-key
OPENAI_API_BASE=https://api.hyperbolic.xyz/v1
OPENAI_MODEL=meta-llama/Meta-Llama-3.1-405B

Usage

Start the server:

pnpm start

This will start the server on stdio, making it available for MCP clients to communicate with.

Docker Usage

Building the Docker Image

docker build -t mcp-openai-complete .

Running the Container

# Run with environment variables
docker run -it --rm \
  -e OPENAI_API_KEY="your-api-key" \
  -e OPENAI_MODEL="gpt-3.5-turbo-instruct" \
  mcp-openai-complete

You can also use a .env file:

# Run with .env file
docker run -it --rm \
  --env-file .env \
  mcp-openai-complete

Parameters for the "complete" tool

prompt (string, required): The text prompt to complete
max_tokens (integer, optional): Maximum tokens to generate, default: 150
temperature (number, optional): Controls randomness (0-1), default: 0.7
top_p (number, optional): Controls diversity via nucleus sampling, default: 1.0
frequency_penalty (number, optional): Decreases repetition of token sequences, default: 0.0
presence_penalty (number, optional): Increases likelihood of talking about new topics, default: 0.0

Development

For development with auto-reloading:

npm run dev

License

MIT