Production-grade feature extraction utility that parses unstructured text and HTML for technical specifications, returning normalized metadata maps.
Single tool with minimal documentation and schema. The tool name 'extract_specs' lacks a strong action verb (should be 'parse_specs' or 'extract_product_specs'). Description is present but generic at 80 characters. Input schema exists but lacks parameter descriptions. No output schema documented. No error handling guidance. The implementation is functional Python code but does not meet MCP production standards.
Parses input string for standard e-commerce specification labels.
Tool name lacks strong action verb. 'extract_specs' is weak; should be 'parse_product_specs' or 'extract_product_specifications' to clearly signal intent before LLM reads description.
Input parameter 'text' has no description in the schema. The rubric requires EVERY parameter to have a description explaining what it controls. LLMs cannot infer that this accepts HTML or plain text without explicit guidance.
No output schema documented. Tool returns a dict with 'extracted_specifications' key, but the structure of that value (dict of strings? list of objects?) is not declared. LLMs cannot plan downstream tool calls without knowing the response shape.
Tool description is only 80 characters and lacks actionable detail. Should explain: what formats it accepts (HTML, plain text, mixed), what spec labels it recognizes (dimensions, weight, material, color, battery, warranty), what it returns (dict of key-value pairs), and when to use it vs manual parsing.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2026-07-28+ | v2 |
No error handling documented. What happens if input is empty? Non-ASCII text? Extremely long strings? No guidance on retry-ability or recovery. Implementation will silently return empty specs dict on no matches, but LLM gets no signal.
Hardcoded target_keys list (dimensions, weight, material, color, battery, warranty) is inflexible and not discoverable. If a product has 'brand', 'model', or 'frequency_response', they are silently ignored. No mechanism for LLM to discover or extend the spec vocabulary.