Dive MCP Host

Dive MCP Host is a language model host service based on the Model Context Protocol (MCP), providing a unified interface to manage and interact with various language models.

Purpose of the Program

Dive MCP Host offers the following features:

• A unified language model interaction interface, supporting multiple models (such as OpenAI, Anthropic, Google, etc.)
• Conversation management and persistent storage
• HTTP API and WebSocket support
• Command-line tools for quick testing and interaction
• Support for multi-threaded conversations and user management

This project uses LangChain and LangGraph to build and manage language model workflows, providing a standardized way to interact with different language models.

How to Run

Environment Setup

1. Ensure you have Python 3.12 or higher installed
2. Clone this repository
3. Install dependencies:

# Using pip
pip install -e .

# Or using uv pip
uv pip install -e .

# Or using uv sync (recommended, will respect uv.lock file)
uv sync --frozen

4. Activate the virtual environment

source .venv/bin/activate

Starting the HTTP Service

Use the following command to start the HTTP service:

dive_httpd

The server will start on:

• Host: 0.0.0.0
• Port: 61990
• Log Level: INFO (configurable in the service config)

Using the Command Line Tool

You can use the command line tool for quick testing:

# General chat
dive_cli "Hello"

# Resume a chat with a specific thread
dive_cli -c CHATID "How are you?"

Using in Code

from dive_mcp_host.host.conf import HostConfig
from dive_mcp_host.host import DiveMcpHost

# Initialize configuration
config = HostConfig(...)

# Use async context manager
async with DiveMcpHost(config) as host:
    # Start or resume a conversation
    async with host.chat(thread_id="123") as chat:
        # Send a query and get a response
        async for response in chat.query("Hello, how can you help me today?"):
            print(response)

Development

Install development dependencies:

pip install -e ".[dev]"

or

uv sync --extra dev --frozen

(Optional) Start local PostgreSQL

./scripts/run_pg.sh

Run tests:

pytest

or with uv, (no need to activate enviroment)

uv run --extra dev --frozen pytest

License

This project is licensed under the MIT License. See the LICENSE file for details.