mcp-server

haozhuang0000/mcp-server

3.1

If you are the rightful owner of mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

A Model Context Protocol (MCP) server implementation that integrates LangChain and Milvus vector database for intelligent agent workflows and semantic search capabilities.

MCP Server with LangChain and Milvus

A Model Context Protocol (MCP) server implementation that integrates LangChain and Milvus vector database for intelligent agent workflows and semantic search capabilities.

Features

  • MCP Integration: Full Model Context Protocol support for seamless AI agent communication
  • LangChain Compatibility: Built-in support for LangChain tools and LangGraph agents
  • Flexible Vector Database: Configurable schemas for different document types and use cases
  • Multiple Database Support: Abstract interfaces for vector (Milvus) and tabular (MySQL) databases
  • Dynamic Schema Creation: Support for custom collection schemas with configurable fields
  • Intelligent Search: Automatic filter extraction from natural language queries
  • RESTful API: HTTP-based MCP server with streamable client support

Prerequisites

  • Python 3.12+
  • Conda (recommended for environment management)

Installation

1. Install uv

Linux / Mac

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Install dependencies

uv sync

Quick Start

Here's a basic example of how to use the MCP server with LangChain:

cd src
python main.py

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client
from langgraph.prebuilt import create_react_agent
from langchain_mcp_adapters.tools import load_mcp_tools

async with streamablehttp_client("http://MCP_SERVER_ADDRESS:PORT/mcp/") as (read, write, _):
    async with ClientSession(read, write) as session:
        # Initialize the connection
        await session.initialize()
        
        # Load available tools from MCP server
        tools = await load_mcp_tools(session)
        
        # Create a reactive agent with the tools
        agent = create_react_agent("openai:gpt-4.1", tools)
        
        # Execute a query
        response2 = await agent.ainvoke({
            "messages": """please help find the total revenue for company: Singapore Airlines in year: 2024, you should search from vector database, the collection name is annual_report_0821"""
        })
        
        print(response2['messages'][-1].content)

Configuration

Environment Variables

Create a .env file in your project root:

# MCP Server Configuration
MCP_SERVER_HOST=localhost
MCP_SERVER_PORT=8000

# Vector Database Configuration
EMBEDDING_END_POINT=<FASTAPI_HOSTED_EMBEDDING_MODEL>
EMBEDDING_DIM=4096
DEFAULT_COLLECTION=documents

# Milvus Configuration
MILVUS_URL=localhost
MILVUS_DB_NAME=default
MILVUS_PW=your_milvus_password

# MySQL Configuration
MYSQL_HOST=localhost
MYSQL_PORT=3306
MYSQL_USER=your_mysql_user
MYSQL_PASSWORD=your_mysql_password
MYSQL_DATABASE=your_database_name

# OpenAI Configuration
OPENAI_API_KEY=your_openai_api_key_here

Schema Types

The generalized system supports multiple schema types:

  1. Document Schema (default): General-purpose schema for various document types
  2. Annual Report Schema: Specialized for financial reports with company/year filtering
  3. Custom Schema: User-defined schemas for specific use cases

Using Different Schemas

# Document schema (default)
vector_db = DatabaseFactory.create_vector_db(
    "milvus",
    collection_name="my_documents",
    schema_type="document"
)

# Annual reports schema
vector_db = DatabaseFactory.create_vector_db(
    "milvus",
    collection_name="annual_reports",
    schema_type="annual_report"
)

# Custom schema
custom_schema = SchemaFactory.create_custom_schema(
    "custom_collection",
    fields=[...],  # Your custom fields
    embedding_dim=4096
)
vector_db.set_custom_schema(custom_schema)

Acknowledgments