An MCP server that asks a human when it isn't sure - extraction with a human-review gate.
Three tools with clear, verb-based names and reasonable descriptions (100-150 chars each). All tools have input schemas with Zod validation. However, parameter descriptions are minimal or absent, output schemas are undocumented (responses are JSON strings, not structured objects), and error handling lacks recovery guidance. The server demonstrates a coherent domain pattern (extraction with human review gate) but falls short of production-grade polish in parameter clarity and output structure.
Extract fields from a document. High-confidence values return immediately; uncertain values are held in a human-review queue instead of guessed. Fields: invoice_number, total_amount, date, email, vendor_name.
A human supplies the correct value for a held item, closing the loop.
List extraction items currently waiting for a human decision.
Parameter descriptions missing or incomplete. 'fields' in extract_with_gate lacks detail on valid field names; 'value' in resolve_review has minimal context. LLMs cannot infer constraints from names alone.
Output schemas undocumented. All tools return JSON strings wrapped in text content blocks. LLMs cannot parse expected fields (e.g., 'extracted', 'needs_human', 'resolved') without reading the description. Structured output with typed fields would enable better chaining.
Error handling lacks recovery guidance. resolve_review returns {'error': 'unknown review_id'} as a JSON string with no suggestion for next steps (e.g., 'Call review_queue() to list valid IDs'). Agents cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
No input validation or constraint documentation. extract_with_gate accepts any array of field names; invalid fields silently return null. Parameter descriptions should enumerate valid fields: 'One of: invoice_number, total_amount, date, email, vendor_name.'
review_queue tool name is ambiguous. 'review_queue' could mean 'get_review_queue', 'list_pending_reviews', or 'check_queue_status'. Verb-first naming (list_review_queue or get_pending_reviews) would clarify intent.