MCP Llm Bridge

🚀 MCP LLM Bridge

A bridge connecting the Model Context Protocol (MCP) server with OpenAI-compatible Large Language Models (LLMs). It primarily supports the OpenAI API and is also compatible with local endpoints implementing the OpenAI API specification.

This implementation provides a bidirectional protocol conversion layer that converts between MCP and OpenAI's function call interfaces. It transforms MCP tool specifications into the OpenAI function architecture and handles the process of mapping function calls back to MCP tool execution. This enables any OpenAI-compatible language model to use MCP-compatible tools through a standardized interface, whether it's a cloud-based model or a local implementation (such as Ollama).

For more information about MCP, please refer to the following links:

• Resources
• Prompts
• Tools
• Sampling

🚀 Quick Start

Installation

# Installation
curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/bartolli/mcp-llm-bridge.git
cd mcp-llm-bridge
uv venv
source .venv/bin/activate
uv pip install -e .

# Create a test database
python -m mcp_llm_bridge.create_test_db

📦 Installation

Configuration

OpenAI (Main Support)

Create a .env file:

OPENAI_API_KEY=your_key
OPENAI_MODEL=gpt-4o # Or other OpenAI models supporting tools

⚠️ Important Note

If you need to use the key in .env, reactivate the environment: source .venv/bin/activate

Then configure the bridge in src/mcp_llm_bridge/main.py:

config = BridgeConfig(
    mcp_server_params=StdioServerParameters(
        command="uvx",
        args=["mcp-server-sqlite", "--db-path", "test.db"],
        env=None
    ),
    llm_config=LLMConfig(
        api_key=os.getenv("OPENAI_API_KEY"),
        model=os.getenv("OPENAI_MODEL", "gpt-4o"),
        base_url=None
    )
)

Additional Endpoint Support

The bridge also works with any endpoint implementing the OpenAI API specification:

Ollama

llm_config=LLMConfig(
    api_key="Not required",
    model="mistral-nemo:12b-instruct-2407-q8_0",
    base_url="http://localhost:11434/v1"
)

💡 Usage Tip

After testing, I found that mistral-nemo:12b-instruct-2407-q8_0 is better at handling complex queries.

LM Studio

llm_config=LLMConfig(
    api_key="Not required",
    model="local-model",
    base_url="http://localhost:1234/v1"
)

I haven't tested this configuration, but it should theoretically work.

💻 Usage Examples

Basic Usage

python -m mcp_llm_bridge.main

# Example question: What is the most expensive product in the database?
# Enter 'quit' or press Ctrl+C to exit

🧪 Running Tests

Install the package with test dependencies:

uv pip install -e ".[test]"

Then run the tests:

python -m pytest -v tests/

📄 License

MIT

🤝 Contribution Guidelines

If you'd like to contribute, please refer to the project repository: https://github.com/bartolli/mcp-llm-bridge