Google Knowledge Graph MCP

🚀 Google Knowledge Graph Search MCP

A Model Context Protocol server that connects Claude (or any MCP client) to Google's free public Knowledge Graph API, enabling searches for real - world entities and retrieval of structured data.

🚀 Quick Start

This is an MCP server. If you're unfamiliar with MCP, you may not need this. However, if you're using Claude Desktop or another MCP - compatible client and wish to search Google's knowledge database, this is the right tool for you.

✨ Features

Why This Exists

I developed this to allow Claude to verify entity information during research workflows. Google's Knowledge Graph contains structured data about millions of real - world entities, the same data that powers the knowledge panels in Google Search results. The Knowledge Graph Search API is completely free, requiring only a Google Cloud API key without any billing account or usage costs.

What You Get

There are two tools for searching Google's knowledge graph:

• Search by query - search_knowledge_graph: Search for entities by name or description. It returns structured data including entity types, descriptions, Wikipedia URLs, and relevance scores.
• Lookup by MID - lookup_knowledge_graph_entities: If you already have Machine IDs (Google's internal entity identifiers), you can look them up directly, which is useful for entity resolution workflows.

Both tools return JSON with entity names and types, detailed descriptions (usually from Wikipedia), official images and URLs, result scores (relevance ranking), and Machine IDs for further lookups.

📦 Installation

NPX (Easiest)

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "google-knowledge-graph": {
      "command": "npx",
      "args": ["-y", "@houtini/google-knowledge-graph-mcp"],
      "env": {
        "GOOGLE_KNOWLEDGE_GRAPH_API_KEY": "your-api-key-here"
      }
    }
  }
}

Local Install

If you prefer running from source:

git clone https://github.com/houtini/google-knowledge-graph-mcp.git
cd google-knowledge-graph-mcp
npm install
npm run build

Then configure Claude Desktop:

{
  "mcpServers": {
    "google-knowledge-graph": {
      "command": "node",
      "args": ["C:\\path\\to\\google-knowledge-graph-mcp\\dist\\index.js"],
      "env": {
        "GOOGLE_KNOWLEDGE_GRAPH_API_KEY": "your-api-key-here"
      }
    }
  }
}

Windows users: Use double backslashes in paths: C:\\MCP\\...

Config location:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

💻 Usage Examples

Once installed and Claude Desktop is restarted, you can:

Basic Usage

Search the knowledge graph for "Marie Curie"

Advanced Usage

• Entity type filtering:

Search knowledge graph for "Python" with types ["ComputerLanguage"]

• Multiple results:

Search knowledge graph for "Paris" limit 5

• Lookup by MID:

Look up knowledge graph entity /m/0dl567

The MCP returns structured JSON that Claude can parse, including entity names, types, descriptions, URLs, and relevance scores.

📚 Documentation

What Gets Returned

Example response structure:

{
  "entities": [
    {
      "mid": "/m/0dl567",
      "name": "Taylor Swift",
      "type": ["Person", "Thing"],
      "description": "American singer-songwriter",
      "detailedDescription": "Taylor Alison Swift is an American singer-songwriter...",
      "image": "https://...",
      "url": "http://en.wikipedia.org/wiki/Taylor_Swift",
      "resultScore": 4258.07
    }
  ],
  "count": 1
}

Parameters

search_knowledge_graph

• query (required): Search term
• languages (optional): Language codes array, e.g., ["en"]
• types (optional): Entity types to filter by, e.g., ["Person", "Organization"]
• limit (optional): Max results (default 20, max 500)

lookup_knowledge_graph_entities

• ids (required): Array of Machine IDs (MIDs), e.g., ["/m/0dl567"]
• languages (optional): Language codes array

Common Entity Types

The Knowledge Graph uses schema.org types. Common ones include:

• Person - Individual people
• Organization - Companies, institutions
• Place - Locations, geographical entities
• Event - Historical or current events
• CreativeWork - Books, films, music, art
• Product - Commercial products
• ComputerLanguage - Programming languages
• SportsTeam - Sports teams
• Country - Nations and countries
• City - Cities and municipalities

You can combine types for more specific searches.

🔧 Technical Details

• API Endpoint: https://kgsearch.googleapis.com/v1/entities:search
• Response Format: JSON - LD with itemListElement array
• Authentication: API key via query parameter
• Rate Limits: Free tier quotas apply (generally 100,000 queries/day)
• Module Format: CommonJS (compatible with Node.js MCP hosts)

Why CommonJS?

The MCP SDK uses CommonJS patterns. I've stuck with that for compatibility. If you're building your own MCP and want ES modules, that's fine - just different choices.

📄 License

This project is licensed under the MIT License. You can do what you want with it.

Troubleshooting

MCP not appearing in Claude

1. Check your JSON syntax - one error breaks everything.
2. Verify the path uses correct escaping (\\ on Windows).
3. Completely restart Claude Desktop (quit, not just minimise).
4. Check the API key environment variable spelling.

"API key required" error

• The environment variable isn't being read.
• Check spelling: GOOGLE_KNOWLEDGE_GRAPH_API_KEY.
• Restart Claude Desktop after config changes.

No results returned

• Try different query terms.
• Remove entity type filters to broaden search.
• Check result limit isn't set too low.

401 Unauthorized

• API key is invalid or expired.
• Knowledge Graph Search API isn't enabled in your Google Cloud project.

Building From Source

npm install
npm run build

The build process compiles TypeScript to CommonJS in dist/. No special configuration needed.

Contributing

If you find issues or have improvements:

1. Check existing issues first.
2. Test your changes locally.
3. Submit a PR with clear description.

I'm particularly interested in hearing about:

• Entity types that need better handling.
• Response parsing edge cases.
• Real - world usage patterns.

Author

Built by Richard Baxter (Houtini) as part of a collection of MCP servers for AI - assisted development and research workflows.

Other Houtini MCPs:

• @houtini/gemini-mcp - Google AI chat with grounding and deep research
• @houtini/geo-analyzer - Content optimisation for AI search engines
• @houtini/brevo-mcp - Email marketing automation

Related

• Model Context Protocol - Protocol specification
• Google Knowledge Graph API - Official API docs
• Claude Desktop - Primary MCP client

Version: 1.0.0
Status: Production ready, tested with Claude Desktop