patents-mcp

An MCP (Model Context Protocol) server for patent search and prior art discovery, powered by the Google Patents public dataset on BigQuery.

Built with the official Go MCP SDK.

Why patents-mcp?

Existing patent MCP servers are either paid, single-source, or unmaintained. patents-mcp is:

• Free — built on Google Patents public dataset (1TB/month free on BigQuery)

• Open source — MIT licensed

• Production grade — structured logging, graceful shutdown, clean architecture

• Extensible — each tool is self-contained, easy to add new tools

Available Tools

Prerequisites

• Go 1.21+

• A Google Cloud account (free)

• BigQuery API enabled on your GCP project

Setup

1. Google Cloud Setup

1. Create a GCP project at console.cloud.google.com

2. Enable the BigQuery API

3. Create a service account with the following roles:

   • BigQuery Job User

   • BigQuery Data Viewer

4. Download the service account JSON key

2. Install

git clone https://github.com/tushariitr-19/patents-mcp
cd patents-mcp
go build -o patents-mcp-server ./cmd/server/

3. Configure

export GOOGLE_APPLICATION_CREDENTIALS="/path/to/service-account.json"
export GCP_PROJECT_ID="your-gcp-project-id"

# Optional: enable debug logging
export DEBUG=true

Usage with Claude Desktop

Add to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "patents-mcp": {
      "command": "/path/to/patents-mcp-server",
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json",
        "GCP_PROJECT_ID": "your-gcp-project-id"
      }
    }
  }
}

Example Prompts

Once connected to Claude Desktop:

• "Search for patents related to transformer neural networks"

• "Find prior art for context-aware UI element hiding"

• "What patents has Google filed related to quantum computing?"

• "Find patents by inventor Yann LeCun"

Architecture

patents-mcp/
├── cmd/server/main.go       ← entry point, env vars, graceful shutdown
├── server/server.go         ← MCP server setup, tool registration
├── tools/
│   └── search.go            ← search_patents tool (self-contained)
├── bigquery/
│   └── client.go            ← BigQuery query execution
├── logger/
│   └── logger.go            ← structured logging via zap
└── models/
    └── patent.go            ← shared Patent struct

Each tool owns its own dependencies — the server is agnostic of what tools do internally. Adding a new tool is a single line in server/server.go.

Contributing

PRs welcome. To add a new tool:

1. Create tools/<toolname>.go

2. Define your input struct and tool definition

3. Register it in server/server.go with one line

License

MIT