Non-generic Deep Research: search (no API key), parallel scrape 3-8 sources, synthesize with citations, fact-check, compare sources, persistent memory. 8 tools.
Server has 8 tools with mostly complete schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (search, scrape, deep_research, fact_check, compare_sources, save_memory, recall_memory, get_research_history). Descriptions range 50-180 chars, meeting baseline minimums. However, parameter descriptions are sparse or missing context. Output schemas are not documented, LLMs cannot predict return structure. Error handling is absent from tool definitions. No tool annotations (readOnlyHint/destructiveHint). The deep_research and compare_sources tools combine multiple concerns (search + scrape + synthesize), violating single-responsibility. Memory operations (save_memory, recall_memory) lack transaction semantics or conflict resolution guidance.
Compare multiple sources on a topic to identify agreements, disagreements, and unique perspectives.
Perform deep research on a topic: search, scrape multiple sources in parallel, synthesize findings with citations, and save to persistent memory.
Fact-check a specific claim by searching for supporting/contradicting evidence and comparing sources.
Retrieve the history of research, searches, fact-checks, and comparisons performed.
Retrieve saved memories by key or tag from persistent storage.
Save a research finding or insight to persistent memory with tags for later retrieval.
Scrape a single webpage and extract structured content including title, description, text, markdown, headings, and links. Includes SSRF protection.
Output schemas not documented. LLMs cannot predict return structure for any tool. deep_research, compare_sources, and fact_check_claim return synthesized findings but no schema is visible in code or mcp.json.
deep_research and compare_sources combine multiple responsibilities (search + scrape + synthesize). Should split into separate tools: search_and_scrape, synthesize_findings, compare_sources_analysis.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). save_memory is destructive/write but unmarked. search, scrape_page, recall_memory are read-only but unmarked. Agents cannot reason about side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Search the web using DuckDuckGo (no API key required). Returns title, URL, snippet, and source for each result.
Parameter descriptions lack context. 'depth' enum in deep_research has no explanation of what 'quick', 'standard', 'deep' mean in terms of time/sources/synthesis. 'extractMainOnly' in scrape_page lacks guidance on when to use.
No error handling guidance in tool definitions. If search fails, scrape times out, or memory write fails, LLM has no recovery path. No retryable vs fatal classification.