Loading ToolBench
Score code for patentability inside Cursor, Claude, and Codex — keyless MCP with prior-art search, USPTO pillar scores, and patent lookup. Hosted at patentprecheck.com/mcp.
Patent PreCheck MCP has 12 well-named tools with clear descriptions (avg 180 chars, within baseline 194). All tools start with action verbs (precheck_*). Input schemas are present and mostly complete with type definitions and enums. However, output schemas are NOT documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions are adequate but lack format/constraint details (e.g., 'report_id' format PPC-YYYY-MM-DD-XXXXX is mentioned in description but not in schema). Error handling is minimal, no recovery guidance or categorization. Tools like precheck_start_review and precheck_session_status accept optional parameters but lack clear dependency documentation. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.
Compare invention text to a known US patent: embedding similarity and prior-art risk analysis.
Suggest Cooperative Patent Classification (CPC) codes for an invention description. Offline heuristic — informational only.
Return download URLs for finalized ICR deliverables (filing packet, coaching report, package zip, scorecard PDF).
Return a short snippet of current US software-patent legal guidance relevant to this invention.
Resolve a US patent or application number via USPTO ODP and optional corpus join.
List the patentability pillars and band rules used by precheck_score. No network call.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nested objects) for downstream tool chaining or data extraction. Tools return rendered text only, no structured JSON schema for responses.
No tool annotations despite all 12 tools being read-only. Missing readOnlyHint metadata prevents agents from understanding safety profile and optimizing retry/caching strategies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Return the closest prior-art matches (titles, sources, similarity, URLs) for an invention.
Preview examination-risk signals and similar office-action / abandonment patterns.
Run a patentability pre-check on source code or an invention description. Returns pillar scores, band, opportunities, prior-art count, rejection patterns, and legal context when available. Provide `code` or `path`.
Fast semantic search against the 2M+ prior-art corpus without LLM scoring. Returns ranked matches with similarity scores.
Return status for an active Interactive Code Review session. Requires report_id and session_key.
Return the URL to start an Interactive Code Review (optionally with a promo code).
Parameter format constraints not in schema. 'report_id' format (PPC-YYYY-MM-DD-XXXXX) and 'session_key' format are mentioned in descriptions but not enforced via JSON Schema pattern or format fields. LLMs cannot validate before calling.
No error recovery guidance. Tools lack actionable error messages. E.g., HTTP 402 'Upgrade' error hints at payment but does not guide agent on next steps (call precheck_start_review, provide promo code, etc.).
Parameter dependencies undocumented. precheck_start_review accepts optional 'report_id' and 'promo' but does not clarify: is report_id required? Can promo override payment? What happens if both are missing?