Patent PreCheck MCP — Score Code in Cursor & Claude

Find us in MCP directories

Already listed on Glama and Smithery, plus the official MCP Registry (io.github.Patent-PreCheck/patent-precheck) and npm (@patentprecheck/mcp). Same 12 tools whether you install locally or point at the hosted URL.

Fastest: hosted MCP (nothing to install)

Point your AI assistant at our server. Same 12 tools, same 1,500,000+ prior-art corpus — invention text only (no file paths). No API keys.

// Cursor — .cursor/mcp.json (copy-paste, then reload)
{ "mcpServers": { "patent-precheck": { "url": "https://patentprecheck.com/mcp" } } }

Then ask: “Run precheck_score on this module” or “Is this patentable? Use precheck_prior_art.”

Local MCP / CLI (reads files on your machine)

New to these terms? Here’s the plain version: a CLI is a command you type on your computer; an MCP server is a small helper that lets your AI assistant (Claude, Cursor, ChatGPT/Codex, etc.) run that check for you; an IDE is the editor you build in, like Cursor. You add one line once — then you can just ask your AI tool to “run a patent precheck on this.”

One free install gives you both: a precheck command you can type yourself, and the helper your AI assistant can call for you. No keys to set up — the scoring engine and prior-art library stay on our servers.

# Run instantly, no install
npx -y @patentprecheck/mcp score ./src/core/invention.ts

# Or install globally
npm i -g @patentprecheck/mcp
precheck score ./src/core/invention.ts --min-score 60   # exit 4 below threshold (CI gate)
cat invention.md | precheck score -                     # read from stdin
precheck pillars                                         # scoring reference (no network)

Use Cursor? Add it in one click — no file editing:

Prefer to set it up by hand, or using another AI tool? Add this so the assistant can score the code it just wrote, without leaving the editor:

// Cursor — .cursor/mcp.json
{ "mcpServers": { "patent-precheck": { "command": "npx", "args": ["-y", "@patentprecheck/mcp", "mcp"] } } }

# Claude Code
claude mcp add patent-precheck -- npx -y @patentprecheck/mcp mcp

MCP tools exposed to the agent (12 total):

Tool	What it does
`precheck_score`	Full patentability score + USPTO pillar bands
`precheck_prior_art`	Closest prior-art matches from our corpus
`precheck_search_corpus`	Fast semantic search (no LLM scoring)
`precheck_lookup_patent`	Resolve a US patent number (e.g. US1234567B2)
`precheck_compare_to_patent`	Your invention vs. a known patent
`precheck_rejection_patterns`	Examination-risk preview
`precheck_legal_context`	Current US software-patent guidance
`precheck_pillars`	Scoring reference (offline)
`precheck_start_review`	Start Interactive Code Review (filing package)
`precheck_cpc_suggest`	CPC classification hints (offline)
`precheck_session_status`	ICR session status
`precheck_deliverables`	Download links for filing deliverables

Not another USPTO lookup bot. Dozens of Glama MCPs fetch patent PDFs and prosecution history — you bring the API key. Patent PreCheck scores the code you just wrote against a 1,500,000+ indexed corpus. Ask your agent: “Can this be patented?”

npm: @patentprecheck/mcp · Registry: io.github.Patent-PreCheck/patent-precheck · Source: GitHub mirror

Chrome extension

The extension is for inventors who build with AI and need a local session estimate against the five USPTO pillars, plus a dated invention journal — not just accepting model output. When you are ready, Get full score (click + confirm) sends journal + human messages for a free calibrated score; nothing uploads until you confirm.

Why install?

Five USPTO pillars in the panel — Subject matter, utility, novelty, non-obviousness, and disclosure as a session estimate while you chat.
AI vs Human capture strip — Always visible counts of your turns vs the model (supporting context, not the headline score).
On-demand calibrated score — Get full score posts only after you confirm; Continue with Interactive Code Review opens the site with your invention text carried forward.
Private by default — Evidence stays in your browser until you explicitly confirm a score or continue to ICR.

How to start

Keep Chrome Enhanced Safe Browsing on (chrome://settings/security).
Install only from the official Chrome Web Store listing (button below) — not from a ZIP or unpacked folder.
Open ChatGPT, Claude, or Gemini — the panel appears automatically.
Log disclosure evidence when prompted; use Get full score for a calibrated result, or open the dashboard to export / continue with ICR.

Add to Chrome — Chrome Web Store Privacy policy

What Chrome’s install prompts mean

Not a whole-browser spy. Chrome’s generic line “Read and change your data on a number of websites” sounds scary, but Patent PreCheck only runs on named AI chat sites you enable (ChatGPT, Claude, Gemini at install; Cursor, Copilot, Replit, and others when you opt in from the toolbar). It reads chat text on those pages to score your human contributions and keep a local invention journal. It does not scan unrelated sites, email, or banking pages.

Keep Enhanced Safe Browsing on. Do not switch Chrome to Standard protection to install Patent PreCheck. Install only from our Chrome Web Store listing. If Enhanced protection shows a temporary “not trusted by Enhanced Safe Browsing” note for a newer publisher, that is Google’s reputation clock — not a malware finding. You can still choose Continue to install from the Store while leaving Enhanced Safe Browsing enabled. Sideloading a ZIP or unpacked build is for developers only and will look less trusted.

Publisher trust checklist (operator): keep the verified publisher badge, link patentprecheck.com as the official homepage in the Chrome Web Store console, keep the privacy policy accurate, never transfer the listing to a new developer account (that resets trust), and stay policy-compliant so Google can grant Enhanced Safe Browsing “trusted” status over time.

Developers: package for CWS upload with ./scripts/package-chrome-extension.sh → dist/patent-precheck-chrome-extension-v<version>.zip. Load unpacked only for local development — never as the public install path. See docs/chrome-web-store-submission.md.

Import from GitHub (browser)

On the free analyzer, choose Import from GitHub to connect any GitHub account (public or private repos), pick your invention files, and Patent PreCheck imports and runs your free score automatically. The free screen may cap how many files are fetched; upgrade keeps that screened bundle (it does not re-fetch omitted files at checkout). After upgrade, use GitHub re-import or review resync for a fuller selection (subject to size limits). See our privacy policy for how GitHub fetch and retention work.

GitHub Actions

For public repos, use the downloadable workflow (runs npx @patentprecheck/mcp on pull requests). The composite action below is for Patent-PreCheck org workflows only while the action repo is private.

patent-precheck-workflow.yml · Full badge & Actions docs

Patent-PreCheck org — composite action:

jobs:
  patentability:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: Patent-PreCheck/Patent-PreCheck/.github/actions/patent-precheck@v1
        with:
          path: src
          # min-score: '50'
          tier: free

README badge:

[![Patent PreCheck scored](https://patentprecheck.com/img/badges/patent-precheck-scored.svg)](https://patentprecheck.com/analyze)

Command line

From any project (Node 20+), download the CLI script once:

mkdir -p scripts
curl -fsSL -o scripts/precheck-from-file.js \
  https://patentprecheck.com/downloads/precheck-from-file.js
node scripts/precheck-from-file.js ./src/core/invention.ts
# Env: PPC_API_URL, PPC_TIER=free, PPC_MIN_SCORE (optional)

Output is JSON with patentability_score, filing_readiness_score, and patentability_held_back_by.

Cursor & other IDEs

The MCP server above is the best fit for Cursor, Claude Code, Codex, and Antigravity — once configured, just ask the agent to “run a patent precheck on this file.” Prefer a nudge rule instead? Download cursor-patent-precheck.md into your project’s .cursor/rules/. Good moments to run it:

After adding novel logic: precheck score the module (or let the agent call precheck_score)
Before a major refactor: capture a baseline score of the core algorithm
Before investor diligence: email the free report ID to yourself from the results page

API shape

POST /.netlify/functions/analyze
Content-Type: application/json

{
  "code": "<invention text or source>",
  "filename": "main.ts",
  "tier": "free",
  "ai_assistance_declared": "yes_some"
}

Free tier returns scores, pillar details, and prior art count — not full prior art documents (paid review only).

Try the web analyzer

Score your code for patentability inside Cursor, Claude & ChatGPT