Weaviate MCP Server

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

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