DocReader MCP Client - Complete tasks according to document content. Workflow: 1. Search relevant pages from documentation sites 2. Extract content from specific pages 3. (Optional) Abstract excessive content to avoid redundancy 4. Complete the user's instruction based on the document content
DocReaderMCP has a single tool (read_doc) with a vague, multi-step description that conflates several responsibilities. The tool name does not use a clear action verb, 'read_doc' is ambiguous about what actually happens (crawling? searching? extracting? summarizing?). The input schema is present with type declarations, but parameter descriptions are minimal (under 10 chars for most). The tool combines at least three distinct operations (search, extract, summarize) into one, violating the single-responsibility principle. Output schema is not documented. Error handling is absent from the tool definition, no guidance on what happens if the doc_url is invalid, the crawl times out, or the LLM call fails. The implementation shows network calls and LLM integration but the tool interface does not expose this complexity or guide recovery.
Complete the MCP workflow based on the document URL and user question, and return the final reply.
Tool name 'read_doc' is too generic and does not reflect the actual workflow (crawl → search → extract → summarize). LLMs cannot infer intent from the name alone and will misunderstand when to invoke it.
Tool combines multiple responsibilities (search pages, extract content, summarize with LLM) into one tool. This violates single-responsibility design and forces agents to accept an all-or-nothing workflow.
Parameter descriptions are trivial (e.g. 'Document URL', 'User question', 'Crawl depth', 'Number of documents'), under 20 characters each, providing no guidance on format, constraints, or when to use alternative values.
No output schema documented. The tool description claims it returns 'the final reply' but does not specify the structure, content type, or what fields an LLM should extract.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 32 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 32 | - | v1 |
No error handling guidance. If doc_url is unreachable, the crawl times out, or the LLM call fails, the tool provides no recovery instructions or categorization (retryable vs. user-fixable vs. fatal).
Tool description does not declare what it does, when to use it, or what it returns, it just names the workflow. LLMs need explicit context to make selection decisions.
The 'depth' and 'max_results' parameters have defaults (2 and 3) but no constraint documentation (min/max, rationale). Unbounded crawl depth could exhaust resources; LLMs cannot reason about safe values.