Web Search MCP Server with ChromaDB Vector Database

This MCP server provides tools for web search and vector database functionality using LangChain and ChromaDB.

Features

Web Search

• Search documentation for popular libraries (LangChain, LlamaIndex, OpenAI)
• Extract content from web pages

Vector Database (ChromaDB)

• Store and retrieve documents with vector embeddings
• Perform semantic similarity search
• Filter documents based on metadata
• Batch operations for efficiency

Setup

1. Install dependencies:

pip install -e .
# or
uv pip install -e .

2. Create a .env file with the following variables:

# Serper API for web search
USER_AGENT=Mozilla/5.0
SERPER_API_URL=https://google.serper.dev/search
SERPER_API_KEY=your_serper_api_key

# ChromaDB configuration
CHROMA_PERSIST_DIRECTORY=./chroma_db
EMBEDDING_MODEL_NAME=sentence-transformers/all-MiniLM-L6-v2

# Transport mode (stdio or sse)
TRANSPORT=stdio

3. Run the server:

python main.py

Available Tools

Web Search

• get_docs(query: str, library: str): Search documentation for specified libraries

Vector Database (ChromaDB)

• add_document_to_vectordb(content: str, metadata: Optional[Dict[str, Any]]): Add a single document to ChromaDB
• search_vectordb(query: str, top_k: int, filter_criteria: Optional[Dict[str, Any]]): Search the vector database
• delete_document_from_vectordb(document_id: str): Delete a document by ID
• batch_add_documents_to_vectordb(documents: List[Dict[str, Any]]): Add multiple documents in a batch
• create_retriever(search_type: str, search_kwargs: Optional[Dict[str, Any]]): Create a retriever for the vector database

Example Usage

# Add a document to the vector database
doc_id = await add_document_to_vectordb(
    content="This is a sample document about ChromaDB vector databases.",
    metadata={"source": "example", "category": "vector_db"}
)

# Search for similar documents
results = await search_vectordb(
    query="How do vector databases work?",
    top_k=2,
    filter_criteria={"category": "vector_db"}
)

# Delete a document
status = await delete_document_from_vectordb(document_id=doc_id)

# Add multiple documents at once
doc_ids = await batch_add_documents_to_vectordb([
    {
        "content": "Document 1 content",
        "metadata": {"source": "example", "category": "general"}
    },
    {
        "content": "Document 2 content",
        "metadata": {"source": "example", "category": "specific"}
    }
])

Development

• Format code: black . and isort .
• Lint code: ruff check .
• Type check: mypy .