Loading ToolBench
Prior-art exploration for software projects: an MCP server and Agent Skill that find related patents and report observations, never an infringement verdict
Patent-checker demonstrates solid tool design with consistent naming (verb_noun pattern), complete input schemas across all 13 tools, and clear descriptions. All tools are READ_ONLY and properly scoped. However, output schemas are not documented in the visible code, parameter descriptions lack detail on constraints/formats, and error handling guidance is absent. Tool names are action-oriented (ops_search, get_biblio, verify_batch) and descriptions are adequate (avg ~80 chars), meeting baseline expectations. The service layer comment indicates validation happens, but specific error messages and recovery guidance are not visible in tool definitions.
Deduplicate patent families from a batch of publications
Retrieve bibliographic data for a patent publication
Retrieve claims for a patent publication
Retrieve patent family information for a publication
Retrieve legal status events for a patent publication
Normalize and validate a publication number
Search EPO OPS using CQL expressions
Output schemas not documented. Tool descriptions state what is returned (e.g., 'bibliographic data', 'legal status events') but do not specify field names, types, or structure. LLMs cannot plan downstream tool calls or extract specific fields without knowing response shape.
Parameter descriptions lack constraint details. 'CQL search expression (max 4000 characters)' is good, but most batch parameters only say 'List of X' without specifying max element count, element shape, or total payload size. Agents cannot validate inputs before calling.
No error handling guidance visible. Tool definitions do not indicate what errors are possible, which are retryable, or what the LLM should do if a call fails. Comments mention validation but no error messages or recovery paths are documented.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | <=2025-11-25 | v2 |
Search EPO OPS and return bibliographic data
Execute a search plan with multiple CQL queries
Report the current server status and configuration
Generate a usage report of API calls and cache statistics
Verify a batch of publication numbers
Check watch list of publications for updates
Pagination not visible in tool definitions. ops_search and ops_search_biblio accept 'begin' and 'end' parameters but do not document total count, next_cursor, or result limits. Large result sets could exhaust context windows.
Tool composition risk: ops_search and ops_search_biblio appear to do similar things (search EPO OPS) but return different data. Naming does not make the distinction obvious, LLMs may conflate them or pick the wrong one.