alloy-automation-RAG

lashineLaw/alloy-automation-RAG

3.2

If you are the rightful owner of alloy-automation-RAG and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

This document provides a comprehensive guide to configuring and using the Zilliz Cloud MCP server with Claude Code for vector database operations.

Tools
4
Resources
0
Prompts
0

Zilliz Cloud MCP Server Configuration

This directory contains configuration for the Zilliz Cloud MCP (Model Context Protocol) server integration with Claude Code.

🎯 Overview

Zilliz Cloud is a fully managed vector database service (managed Milvus). This MCP server enables Claude Code to interact with your vector database using natural language commands.

Cluster: lastlora-vector Region: GCP US West 1 (Serverless) MCP Package: @zilliz/claude-context-mcp

🚀 Quick Start

Prerequisites

  • Node.js installed
  • Claude Code CLI or Claude Desktop
  • Zilliz Cloud account with active cluster

Installation

The MCP server is already installed globally:

npm install -g @zilliz/claude-context-mcp

Configuration

Credentials are stored in .env.local (not committed to git):

# Load environment variables
source .env.local  # Linux/Mac
# or
export $(cat .env.local | xargs)  # Alternative

🔑 Credentials

All credentials are stored in .env.local file in this directory. Never commit this file to version control.

Required Credentials

  • ZILLIZ_API_KEY: API authentication key
  • ZILLIZ_TOKEN: Database access token
  • ZILLIZ_ENDPOINT: Cluster endpoint URL

Optional Credentials

  • ZILLIZ_USER: Database username
  • ZILLIZ_PASSWORD: Database password
  • ZILLIZ_ORG_ID: Organization identifier
  • ZILLIZ_CLUSTER_ID: Cluster identifier
  • ZILLIZ_CLUSTER_NAME: Human-readable cluster name

💻 Usage with Claude Code

MCP Server Tools

Once configured, you can use natural language commands like:

Collection Management
"List all collections in my Zilliz database"
"Create a collection called 'product_embeddings' with 1536 dimensions"
"Describe the 'documents' collection"
"Delete the 'old_data' collection"
Vector Operations
"Insert vectors with metadata into the 'embeddings' collection"
"Search for the top 5 most similar vectors to [query_vector]"
"Delete vectors where id in [1, 2, 3]"
Hybrid Search
"Search for vectors similar to [query] where category='real-estate' and price > 500000"
"Find documents similar to [text] published after 2024-01-01"
Cluster Information
"Show cluster statistics"
"Get storage usage information"
"List all indexes in the database"

🏗️ Cluster Details

Cluster Information

  • Name: lastlora-vector
  • ID: in05-520fac19203081e
  • Type: Serverless (auto-scaling)
  • Region: GCP US West 1
  • Provider: Google Cloud Platform

Endpoint

https://in05-520fac19203081e.serverless.gcp-us-west1.cloud.zilliz.com

Features

  • ✅ Serverless architecture (auto-scaling)
  • ✅ Full text search (Milvus 2.5+)
  • ✅ Hybrid search (vector + metadata)
  • ✅ Multiple similarity metrics (L2, IP, COSINE)
  • ✅ Dynamic index optimization

📚 Common Use Cases

1. Semantic Search

Store and search document embeddings for semantic similarity matching.

2. Image Similarity

Store image embeddings and find visually similar images.

3. Recommendation Systems

Store user/product embeddings for personalized recommendations.

4. Question Answering

Store knowledge base embeddings for RAG (Retrieval Augmented Generation).

5. Anomaly Detection

Store normal patterns and identify anomalies via vector distance.

🔧 Configuration Files

.env.local

Contains all sensitive credentials. EXCLUDED FROM GIT.

CLAUDE.md

Project-specific guidance for Claude Code when working with Zilliz operations.

.gitignore

Ensures sensitive files are not committed to version control.

🛡️ Security Best Practices

DO ✅

  • Store credentials in .env.local
  • Add .env.local to .gitignore
  • Rotate API keys regularly
  • Use environment variables in scripts
  • Limit API key permissions to minimum required

DON'T ❌

  • Commit .env.local to git
  • Share credentials in plain text
  • Hardcode credentials in source files
  • Use production credentials in development
  • Expose credentials in logs or error messages

📖 Documentation

Official Resources

GitHub Repositories

Tutorials

🧪 Testing the Connection

Quick Test

Load environment variables and test connection:

# Load credentials
export $(cat .env.local | grep -v '^#' | xargs)

# Test connection (via Claude Code MCP)
# In Claude Code, ask:
"List all collections in my Zilliz database"

Expected Response

You should see a list of collections in your lastlora-vector cluster.

🔍 Troubleshooting

Connection Issues

Problem: "Failed to connect to Zilliz Cloud"

Solutions:

  1. Verify .env.local credentials are correct
  2. Check cluster is active in Zilliz Cloud console
  3. Verify endpoint URL is accessible
  4. Ensure API key has proper permissions

Authentication Issues

Problem: "Invalid API key or token"

Solutions:

  1. Regenerate API key in Zilliz Cloud console
  2. Update .env.local with new credentials
  3. Reload environment variables
  4. Restart Claude Code if needed

MCP Server Issues

Problem: "MCP server not responding"

Solutions:

  1. Verify @zilliz/claude-context-mcp is installed globally
  2. Check Node.js version compatibility
  3. Restart Claude Code
  4. Check MCP server logs for errors

📊 Cluster Status

To check your cluster status:

  1. Visit https://cloud.zilliz.com/
  2. Navigate to your organization: org-lseftrpmtsrqnczohrvsts
  3. View cluster: lastlora-vector
  4. Monitor usage, storage, and performance

🎓 Example Workflows

Workflow 1: Create and Populate Collection

1. "Create a collection called 'property_embeddings' with 1536 dimensions using COSINE similarity"
2. "Insert vectors with property metadata (address, price, sqft) into property_embeddings"
3. "Search for properties similar to [embedding_vector]"
4. "Get statistics for the property_embeddings collection"

Workflow 2: Hybrid Search

1. "Search property_embeddings for vectors similar to [query] where price < 2000000 and city='San Francisco'"
2. "Find top 10 most similar properties with at least 3 bedrooms"

Workflow 3: Collection Management

1. "List all collections"
2. "Describe the 'documents' collection schema"
3. "Get row count for all collections"
4. "Drop the 'old_embeddings' collection"

🔗 Related Projects

If you're working with embeddings, consider:

  • OpenAI Embeddings: text-embedding-3-small (1536 dims)
  • Cohere Embeddings: embed-english-v3.0 (1024 dims)
  • Voyage AI: voyage-2 (1024 dims)
  • Custom Models: Any model that outputs vector embeddings

📝 Notes

  • Serverless clusters auto-scale based on usage
  • Free tier available for testing and development
  • Production workloads may require capacity unit reservations
  • Monitor usage to avoid unexpected costs
  • Backup important collections regularly

Created: 2025-10-25 Last Updated: 2025-10-25 Cluster: lastlora-vector (in05-520fac19203081e) Status: ✅ Configured and ready for use