patent-fetch

Patent Fetch

Fetch detailed patent information by patent ID from Google Patents using the google-patent-cli MCP server.

Purpose

Retrieve complete patent details including title, abstract, description, claims, assignee, filing dates, and legal status.

MCP Tool

Uses fetch_patent MCP tool provided by google-patent-cli.

Usage

Fetch a patent, then use the returned dataset name to query with Cypher:

patent_fetch({
  patent_id: "US9152718B2"
})

# Returns dataset name like "fetch-abc123"
# IMPORTANT: Always use execute_cypher to get the patent data
# Do NOT read the output_file directly with Read tool

Response Guidelines

Default behavior (when user doesn't specify what they need): Return ONLY title and abstract_text:

MATCH (p:Patent) RETURN p.title, p.abstract_text

When user asks for specific information, include relevant fields:

User Request	Fields to Return
Assignee, owner, applicant	p.assignee
Legal status, status, expired/active	p.legal_status
Claims, what is claimed	See claims section below
Description, details, specification	p.description
Filing date, application date	p.filing_date
Publication date, p.publication_date
Priority date	p.priority_date

Claims retrieval:

p.claims is always null. You MUST query :claims child nodes directly:

MATCH (c:claims) RETURN c.number, c.text

CRITICAL — Cypher Parser Limitations:

The Cypher parser has known bugs. Follow these rules strictly:

1. Do NOT use ORDER BY — causes c.text to return expression: null
2. Do NOT use WHERE clauses — WHERE d.text CONTAINS '...' and WHERE c.number = '1' cause parse errors
3. Do NOT use relationship patterns for property access — (p:Patent)-[:claims]->(c:claims) RETURN c.text returns null
4. Do NOT use p.claims — always null, claims are stored as child nodes
5. Do NOT use wrong labels — [:HAS_CHILD]->(c:claim), [:claim]->(c:claim), [:claims]->(c:claim) return empty

Safe query pattern: MATCH (c:claims) RETURN c.number, c.text (direct node match, no ORDER BY, no WHERE)

Example queries based on user request:

User: "Fetch patent US9152718B2" → Use default: MATCH (p:Patent) RETURN p.title, p.abstract_text

User: "Who owns patent US9152718B2?" → Include assignee: MATCH (p:Patent) RETURN p.title, p.assignee

User: "What's the legal status of US9152718B2?" → Include legal_status: MATCH (p:Patent) RETURN p.title, p.legal_status

User: "Get full details for US9152718B2" → Include everything:

MATCH (p:Patent) RETURN p.title, p.abstract_text, p.description, p.assignee, p.filing_date, p.publication_date, p.legal_status

Then get claims separately:

MATCH (c:claims) RETURN c.number, c.text

Important Notes

• The output_file returned by fetch_patent is for debugging purposes only
• Always use execute_cypher to access patent data - do not use the Read tool on the output_file
• The dataset is automatically loaded into memory for efficient querying

Graph Structure

Patent data is loaded as a graph with the following structure:

• Patent node (:Patent) - Main patent with id, title, abstract_text, etc.
• Array fields become relationships:
  • (:Patent)-[:claims]->(:claims) - Claim nodes with number, text
  • (:Patent)-[:description_paragraphs]->(:description_paragraphs) - Description nodes
  • (:Patent)-[:images]->(:images) - Image nodes

Accessing claims (direct node match — do NOT use relationship patterns or ORDER BY/WHERE):

MATCH (c:claims) RETURN c.number, c.text

Parameters

• patent_id (string, required): Patent ID (e.g., "US9152718B2", "JP2023-123456-A")
• language (string, optional): Language/locale for patent pages (ja, en, zh)