Patent Mcp Server

What is Patent Mcp Server?

It is an MCP (Model Context Protocol) server that gives AI agents access to patent data — CN patents with CPC-aware correction plus US/WO global coverage. It runs locally on your machine with no external API, no subscription, and is MIT licensed.

How to use Patent Mcp Server?

Install via pip install deeparchi-patent-mcp, then configure the server in your MCP client (e.g., Claude Desktop, Cursor, Windsurf, Cline) with the command python -m src.server. Optionally enable BigQuery search by setting up a GCP project and environment variables.

Key features of Patent Mcp Server

• CN patents with CPC-aware correction, plus US/WO global coverage.
• Three built-in tools: get_patent, get_patent_claims, search_patents.
• get_patent and get_patent_claims require zero setup or cost.
• search_patents uses BigQuery (optional) with Firecrawl fallback for CN.
• Keyword search scans both English and Chinese abstracts.
• Smart fallback: web scraping first, then BigQuery if available.

Use cases of Patent Mcp Server

• An AI agent summarizing patent claims from a publication number.
• Searching patents by CPC classification, assignee, country, or date range.
• Analyzing novelty using citations with X/Y/A/D prior art markers.
• Conducting company or city-level patent landscape analysis (e.g., assignee="BOE").

FAQ from Patent Mcp Server

What does it do?

It gives AI agents the ability to read patent details, claims, and search over 1.4 billion patents using public data, with special handling for CN patents.

Do I need an API key or subscription?

No. For 80% of use cases (getting patent details and claims) no API key is needed. BigQuery search is optional and free within the 1 TB/month free tier.

How does CN patent search work?

It uses a three-layer approach: BigQuery for primary search, Firecrawl as a web fallback when BigQuery cost-rejects a query, and Google Patents for detail enrichment — all transparent to the user.

What are the runtime requirements?

Python 3.10+, no external server, no credentials to share. The server runs locally and only makes the same HTTP requests a browser would make.

Can multiple users share one instance?

Yes. Start the server with HTTP/SSE by editing run-http.sh.example and setting a port. Team members connect via the SSE URL.