Weaviate MCP Server

A Model Context Protocol (MCP) server that provides seamless integration with Weaviate vector databases. This server focuses on powerful search capabilities including semantic, keyword, and hybrid search, with plans to expand functionality in future releases.

Features

The Weaviate MCP Server currently provides 11 essential tools for interacting with your Weaviate instance:

Connection & Configuration

• get_config - View current Weaviate configuration (with sensitive values masked)
• check_connection - Test connection to your Weaviate instance

Schema & Collection Management

• list_collections - List all available collections in your database
• get_schema - Get detailed schema information for specific collections or all collections
• get_collection_objects - Retrieve objects from collections with pagination support

Search Capabilities (Primary Focus)

• search - Simplified search interface using hybrid search by default
• semantic_search - Vector similarity search using embeddings for semantic matching
• keyword_search - BM25-based keyword search for exact term matching
• hybrid_search - Combined semantic and keyword search with configurable weighting

Multi-Tenancy Support

• is_multi_tenancy_enabled - Check if a collection supports multi-tenancy
• get_tenant_list - List all tenants for a multi-tenant collection

Quick Start

The MCP server is designed to be used with MCP clients like Claude Desktop. It uses uvx for automatic installation and execution - no manual installation required.

Test the server directly:

uvx mcp-weaviate --help

Requirements

• Weaviate instance (local or cloud)
• API keys for embeddings:
  • OpenAI API key (for OpenAI embeddings)
  • Cohere API key (optional, for Cohere embeddings)

Configuration

MCP Settings Configuration

Add the Weaviate MCP server to your MCP settings file (typically claude_desktop_config.json or similar):

Local Weaviate Instance

{
  "mcpServers": {
    "mcp-weaviate": {
      "command": "/path/to/uvx",
      "args": [
        "mcp-weaviate",
        "--connection-type", "local",
        "--host", "localhost",
        "--port", "8080",
        "--grpc-port", "50051",
        "--openai-api-key", "YOUR_OPENAI_API_KEY"
      ]
    }
  }
}

Weaviate Cloud Services

{
  "mcpServers": {
    "mcp-weaviate": {
      "command": "/path/to/uvx",
      "args": [
        "mcp-weaviate",
        "--connection-type", "cloud",
        "--cluster-url", "https://your-cluster.weaviate.network",
        "--api-key", "YOUR_WEAVIATE_API_KEY",
        "--openai-api-key", "YOUR_OPENAI_API_KEY"
      ]
    }
  }
}

Configuration Options

Option	Description	Default	Environment Variable
--transport	Transport protocol: "stdio" or "streamable-http"	stdio	-
--http-host	Host for HTTP transport	0.0.0.0	-
--http-port	Port for HTTP transport	8000	-
--connection-type	Connection type: "local" or "cloud"	required	-
--host	Host for local Weaviate connection	required for local	-
--port	HTTP port for local Weaviate connection	required for local	-
--grpc-port	gRPC port for local Weaviate connection	required for local	-
--cluster-url	Weaviate Cloud Services URL	required for cloud	WEAVIATE_CLUSTER_URL
--api-key	API key for authentication	required for cloud	WEAVIATE_API_KEY
--openai-api-key	OpenAI API key for embeddings	-	OPENAI_API_KEY
--cohere-api-key	Cohere API key for embeddings	-	COHERE_API_KEY
--timeout-init	Initialization timeout (seconds)	30	-
--timeout-query	Query timeout (seconds)	60	-
--timeout-insert	Insert timeout (seconds)	120	-

Remote Deployment

For deploying the MCP server remotely (e.g., on TrueFoundry, Kubernetes, etc.), use the streamable-http transport:

mcp-weaviate --transport streamable-http --http-port 8000 --connection-type cloud

This exposes the server on HTTP port 8000 with a /health endpoint for health checks.

Tool Reference

Search Tools

search

Simplified search interface using hybrid search with balanced defaults (alpha=0.3).

semantic_search

Vector similarity search using embeddings. Best for finding conceptually similar content.

keyword_search

BM25 keyword search for exact term matching. Best for finding specific terms or phrases.

hybrid_search

Combines semantic and keyword search using Reciprocal Rank Fusion (RRF).

• alpha parameter controls the balance:
  • 1.0 = 100% semantic search
  • 0.0 = 100% keyword search
  • 0.5 = equal weight
  • 0.3 = default (30% semantic, 70% keyword)

Collection Management

get_collection_objects

Retrieve objects from a collection with pagination support:

• limit: Maximum number of objects to return
• offset: Number of objects to skip (for pagination)

Multi-Tenancy

All search and retrieval tools support an optional tenant_id parameter for multi-tenant collections.

Roadmap

The Weaviate MCP Server currently focuses on comprehensive search capabilities. Future releases will include:

• Data Management

  • Object creation and updates
  • Batch imports
  • Delete operations

• Advanced Query Features

  • Filtering and where clauses
  • Aggregations
  • GraphQL query support

• Collection Management

  • Create/modify collections
  • Index management
  • Backup and restore operations

• Enhanced Search

  • Generative search (RAG)
  • Question answering
  • Custom ranking functions

• Distribution & Deployment

  • Smithery registry support
  • NPX installation compatibility

Development

Setting up for development

# Clone the repository
git clone https://github.com/yourusername/mcp-weaviate.git
cd mcp-weaviate

# Install dependencies with uv
uv sync

# Install development dependencies
uv sync --dev

# Run tests
uv run pytest

# Run linting
uv run ruff check .

# Run type checking
uv run mypy .

Running locally

Example:

uv run python -m src.main \
  --connection-type cloud \
  --cluster-url https://your-cluster.weaviate.network \
  --api-key YOUR_API_KEY \
  --openai-api-key YOUR_OPENAI_KEY

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE file for details.

Support

For issues, questions, or suggestions, please open an issue on the GitHub repository.