Search MCP Server

🚀 parquet_mcp_server

A powerful MCP (Model Control Protocol) server that provides tools for performing web searches and finding similar content. This server is designed specifically for Claude Desktop.

🚀 Quick Start

A powerful MCP (Model Control Protocol) server offers tools for web searches and finding similar content. It's tailored for Claude Desktop and comes with key features like web search and similarity search. Ideal for apps needing web search and projects finding similar content based on queries.

✨ Features

• Web Search: Conduct web searches and scrape results.
• Similarity Search: Extract relevant information from previous results.

This server is particularly suitable for:

• Applications that require web search functionality.
• Projects that need to find similar content based on search queries.

📦 Installation

Install via Smithery

Automatically install the Parquet MCP Server for Claude Desktop using Smithery:

npx -y @smithery/cli install @DeepSpringAI/parquet_mcp_server --client claude

Clone this repository

git clone ...
cd parquet_mcp_server

Create and activate a virtual environment

uv venv
.venv\Scripts\activate  # On Windows
source .venv/bin/activate  # On macOS/Linux

Install packages

uv pip install -e .

Environment variable configuration

Create a .env file and add the following variables:

EMBEDDING_URL=http://sample-url.com/api/embed  # Vector embedding service address
VECTOR_DB_HOST=host.docker.internal  # Hostname or IP address of the PostgreSQL vector database
VECTOR_DB_PORT=5432  # PostgreSQL port, default is 5432
VECTOR_DB_NAME=vector_db  # Database name
VECTOR_DB_USER=postgres  # Database username
VECTOR_DB_PASSWORD=postgres  # Database password

💻 Usage Examples

Install PostgreSQL and the pgVector extension

1. Create a new database in PostgreSQL and enable the pgVector extension:

CREATE EXTENSION vector;

2. Create a table to store search results:

CREATE TABLE web_search (
    id SERIAL PRIMARY KEY,
    text TEXT,  -- Store the full web page content or text fragments
    metadata JSONB,  -- Contains metadata such as URLs and titles
    embedding VECTOR(1024),  -- Vector embedding corresponding to the text, with a dimension of 1024
    created_at TIMESTAMP DEFAULT NOW()  -- Record creation time
);

Test Guide

Initialize the database

Use the following command to initialize the PostgreSQL database and insert sample data:

psql -h host.docker.internal -U postgres -d vector_db -f init.sql

Search test

Send a search request containing an embedding vector to the server, for example:

POST /api/search HTTP/1.1
Content-Type: application/json

{
    "embedding": [0.1, 0.2, ..., 0.3],  # Vector of length 1024
    "match_threshold": 0.7,  # Similarity threshold, default is 0.7
    "match_count": 5  # Maximum number of returned results, default is 5
}

PostgreSQL Similarity Search Function

To perform vector-based similarity searches in PostgreSQL, use the following SQL function:

-- Create a function for similarity search
CREATE OR REPLACE FUNCTION match_web_search(
  query_embedding vector(1024),  -- Query vector, length must be 1024
  match_threshold float,  # Similarity threshold
  match_count int  # Maximum number of returned results, default is 5
)
RETURNS TABLE (
  id bigint,
  metadata jsonb,
  text TEXT,  -- Return the full text content
  created_at TIMESTAMP,  -- Return the creation timestamp
  similarity float
)
LANGUAGE plpgsql
AS $$
BEGIN
  RETURN QUERY
  SELECT
    web_search.id,
    web_search.metadata,
    web_search.text,
    web_search.created_at,
    1 - (web_search.embedding <=> query_embedding) as similarity
  FROM web_search
  WHERE 1 - (web_search.embedding <=> query_embedding) > match_threshold
  ORDER BY web_search.created_at DESC,  # Sort by time, latest results first
           web_search.embedding <=> query_embedding  # Sort by similarity
  LIMIT match_count;  # Limit the number of returned results
END;
$$;

This function allows vector-based similarity searches in PostgreSQL, returning results that meet the criteria based on the specified threshold and result count limit. Results are sorted by timestamp in descending order and then by similarity.

Backend Code Example

Here is a simple implementation example of the backend service:

from fastapi import FastAPI, HTTPException
import requests
import json
from typing import List, Optional

app = FastAPI()

class SearchResults(BaseModel):
    id: int
    metadata: dict
    text: str
    created_at: datetime
    similarity: float

@app.post("/api/search", response_model=List[SearchResults])
async def search(web_search: WebSearch) -> List[SearchResults]:
    try:
        # Execute a similarity search query
        response = requests.post(
            "http://localhost:8000/api/vector/search",
            json={
                "vector": web_search.embedding,
                "threshold": 0.7,
                "count": 5
            }
        )
        
        if response.status_code != 200:
            raise HTTPException(status_code=500, detail="Search failed")
            
        return response.json()
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

📚 Documentation

Other Notes

• Ensure that the PostgreSQL version is 13 or higher to support the vector data type.
• The pgVector extension must be installed correctly, and the database must have the extension enabled.
• The web search function depends on third - party crawler tools (such as requests and beautifulsoup4). Please ensure that these libraries are installed and configured correctly.
• When using in a production environment, adjust the similarity threshold and the number of returned results according to actual needs.