Score your code's patentability from inside Cursor

Most developers think about patents far too late — usually after the product ships, when re-creating the “how” and “why” behind the code is painful. We built the Patent PreCheck CLI and MCP server so patentability becomes something you can check while you build, like running a linter or a test.

It wraps our hosted patentability engine. You point it at a file or paste a snippet, and it returns a 0–100 score across the USPTO statutory pillars (§101 eligibility, §102 novelty, §103 non-obviousness, §112 documentation), the band you fall into, the pillar holding you back, and how many prior-art references it consulted.

Install in one command

The package is published on npm as @patentprecheck/mcp. It needs Node.js 20+.

# Run it without installing anything:
npx @patentprecheck/mcp score ./src/feature.ts

# Or install the `precheck` command globally:
npm install -g @patentprecheck/mcp

The analysis engine stays server-side, so there are no model keys to configure and nothing to set up beyond the package itself.

Three ways to use it

1. From the command line

Score a file, print the scoring reference, or get a link to start a full review:

precheck score ./src/limiter.ts     # score a file (or use - for stdin)
precheck pillars                    # explain the five pillars and bands
precheck review                     # get the Interactive Code Review link

2. Inside Cursor, Claude Code, or Codex (MCP)

Add it as an MCP server and your AI agent can score the code it just wrote — and explain the result — without you switching tools. In Cursor, add this to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "patent-precheck": {
      "command": "npx",
      "args": ["-y", "@patentprecheck/mcp", "mcp"]
    }
  }
}

The server exposes three tools: precheck_score (score code or a description), precheck_pillars (explain the scoring), and precheck_start_review (hand off to a paid Interactive Code Review). The same config pattern works for any MCP-compatible agent — see the integrations guide for Claude Code and Codex examples.

3. In CI, as a gate

The score command uses typed exit codes, so you can fail a pull request when an invention scores below a threshold:

# exits non-zero if patentability is under 70
npx @patentprecheck/mcp score ./src/feature.ts --min-score 70 --format json

What you get back

A single run returns:

Patentability score (0–100) and the band — Not Ready → Building → Close → File Ready.
Pillar breakdown — §101 eligibility, §102 novelty, §103 non-obviousness, plus a §112 filing-readiness signal.
The pillar holding your band back, so you know exactly what to strengthen first.
Top opportunities to improve, and a count of prior-art matches consulted.

“But I built it with AI — can it even be patented?”

Yes. AI-assisted code can absolutely be patentable. Under current U.S. law, what matters is meaningful human conception — the direction, decisions, and technical choices you made. The pre-check flags where your work reads as a specific technical improvement versus a generic, abstract flow, and the deeper Interactive Code Review coaches you on documenting that conception so it holds up. For the legal background, see Is AI-generated code patentable? and documenting inventorship with AI IDEs.

Privacy

Free scores are analyzed in-session and not stored afterward. The CLI and MCP server send your code to the hosted analyze endpoint over HTTPS for scoring only — see our privacy policy for retention details, and the security page for how stored code is protected if you later start a paid review.

Not legal advice. Patent PreCheck is a preparation tool; a licensed patent attorney should draft and file your claims.

Next steps

Set up the CLI & MCP server · Run a free score in the browser · View on npm · Can my code be patented?

Score your code’s patentability from inside Cursor