Patent Intel

Patent intelligence for AI agents — patent search, USPTO data, patent landscape, technology trends, and semantic prior-art search (pgvector) over USPTO PatentsView data.

Part of the FoundryNet Data Network. Every result carries verifiable provenance so you can confirm it was produced by this server, unaltered. See also: gov-contracts-mcp, brand-intel-mcp.

Connect

• MCP endpoint (Streamable HTTP): https://patent-intel-mcp-production.up.railway.app/mcp
• MCP registry: io.github.FoundryNet/patent-intel-mcp
• Agent card: https://patent-intel-mcp-production.up.railway.app/.well-known/agent-card.json

Claude Desktop / Cursor / Claude Code

claude mcp add --transport http patent-intel https://patent-intel-mcp-production.up.railway.app/mcp

{ "mcpServers": { "patent-intel": { "url": "https://patent-intel-mcp-production.up.railway.app/mcp" } } }

Tools

Tool	Price	What it does
search_patents	$0.01	Search by keyword, assignee, CPC, date, type
patent_detail	free	Full record: claims, citations, inventors
company_patents	$0.01	Portfolio: count, filing velocity, tech areas, recent filings
trending_technology	$0.01	CPC classes ranked by filing volume, with top assignees
prior_art_search	$0.05	Semantic similarity over abstracts (pgvector) — premium
daily_digest	$0.05	Structured daily patent-filing digest
brief_summary	$0.50	Sample slice of the curated daily patent-intel brief
daily_brief	$10	Full curated daily patent-intel brief — premium
mint_info	free	FoundryNet Data Network details

Free tier: 25 paid-tool queries/day per agent (plus free patent_detail + mint_info). Pass agent_id to scope your allowance. After that, the tool returns an HTTP-402 with a payment challenge — settle it, then re-call with the same args plus payment_tx=<reference>. An Authorization: Bearer fnet_… key bypasses the paywall.

How it works

A daily task (~5am PT) ingests patents granted in the last day from the USPTO PatentsView Search API, maps them to a normalized schema, embeds abstracts with fastembed (BAAI/bge-small-en-v1.5, local, no API key), and stores them in a standalone Supabase project. prior_art_search embeds your query and ranks by pgvector cosine similarity. A rolling patent_assignees summary powers company_patents.

Honesty note: claims_count isn't exposed by the PatentsView patent endpoint, so it's left null (a future enrichment hook); citation_count and the rest come straight from PatentsView.

Also on

Smithery: io.github.FoundryNet/patent-intel-mcp

Built by FoundryNet · [email protected]