weaviate-mcp-server-railway by klogins-hash - MCP Server

Weaviate MCP Server Project

This project contains the official Weaviate MCP (Model Context Protocol) server implementation, optimized for Railway deployment.

Project Structure

server doesn't know v3/
├── mcp-server-weaviate/          # Official Weaviate MCP server
│   ├── client/
│   │   ├── mcp-server            # Built server binary
│   │   └── client.go             # Test client
│   ├── .env                      # Environment configuration
│   ├── Dockerfile                # Railway deployment
│   ├── railway.toml              # Railway configuration
│   ├── start-server.sh           # Server startup script
│   └── ...                       # Other project files
└── README.md                     # This file

Prerequisites

Go 1.25+ (installed via Homebrew)
Weaviate instance (local, cloud, or Weaviate Cloud Services)

Configuration

The server is configured via environment variables in .env:

# Railway will automatically set PORT
PORT=8675

# Weaviate connection settings
WEAVIATE_HOST=localhost:8080
WEAVIATE_SCHEME=http

# For Weaviate Cloud Services (recommended for Railway)
# WEAVIATE_HOST=your-cluster.weaviate.network
# WEAVIATE_SCHEME=https
# WEAVIATE_API_KEY=your_api_key_here

Railway Deployment

Quick Deploy

Connect to Railway: Link your GitHub repository to Railway
Set Environment Variables: Configure Weaviate connection in Railway dashboard
Deploy: Railway will automatically build and deploy using the Dockerfile

Environment Variables for Railway

Set these in your Railway project dashboard:

WEAVIATE_HOST=your-cluster.weaviate.network
WEAVIATE_SCHEME=https
WEAVIATE_API_KEY=your_weaviate_api_key

API Endpoints

Once deployed, your Railway app will expose:

GET / - Server information and available endpoints
GET /health - Health check endpoint
GET /tools - Available MCP tools
POST /call - Execute MCP tools

Local Development

1. Start Weaviate (if running locally)

docker run -p 8080:8080 -e QUERY_DEFAULTS_LIMIT=25 -e AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED=true -e PERSISTENCE_DATA_PATH='/var/lib/weaviate' -e DEFAULT_VECTORIZER_MODULE='none' -e ENABLE_MODULES='' -e CLUSTER_HOSTNAME='node1' semitechnologies/weaviate:latest

2. Build and Start the MCP Server

cd "mcp-server-weaviate"
make build
./start-server.sh

The server will be available at http://localhost:8675

Available Tools

weaviate-insert-one: Insert an object into Weaviate
weaviate-query: Retrieve objects from Weaviate with hybrid search

API Usage Examples

Insert Object

curl -X POST http://your-app.railway.app/call \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "weaviate-insert-one",
    "arguments": {
      "collection": "MyCollection",
      "properties": {
        "title": "Example Document",
        "content": "This is example content"
      }
    }
  }'

Query Objects

curl -X POST http://your-app.railway.app/call \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "weaviate-query",
    "arguments": {
      "query": "example content",
      "targetProperties": ["title", "content"]
    }
  }'

Development Commands

Build: make build
Test client: make run-client (requires server to be running)
Docker build: docker build -t weaviate-mcp-server .
Docker run: docker run -p 8675:8675 --env-file .env weaviate-mcp-server

Notes

The server supports both stdio (for local MCP clients) and HTTP (for Railway deployment)
Railway deployment uses HTTP endpoints for better cloud compatibility
For production, use Weaviate Cloud Services for better reliability
Health check endpoint at /health is configured for Railway monitoring