Authenticated MCP server for systematic literature reviews with verified citations. Surfaces your Thoth reviews + per-claim cite_check audits.
Thoth exposes 5 read-only tools via HTTP transport with consistent naming (verb_noun pattern: list_, get_). All tools have descriptions (50-150 chars, meeting baseline 34-392 range). However, input schemas are not visible in the provided source code, only tool names and descriptions appear in lib/mcp/tools/index.ts. Without seeing parameter definitions, types, or output schemas, schema quality cannot be verified. Descriptions are adequate but generic (e.g., 'Read-only. Lists every Thoth systematic-literature-review run...' lacks WHEN/WHY context). No error handling guidance visible. All tools are read-only (low risk), but lack of schema visibility and missing parameter documentation are significant gaps.
Read-only. Returns Thoth's per-claim cite_check verdict (supported/unsupported/unclear) for every cited claim in a completed review, plus aggregate counts and the run's overall faithfulness score. 404 for unowned reviews.
Read-only. Returns the full markdown draft of a completed Thoth review, plus its critic and faithfulness scores. 404 for unowned reviews.
Read-only. For outbound/hybrid v2 reviews, returns the natural-language search queries the discoverer LLM generated from the research question, the provider set the run targeted, and any per-provider error messages. Empty queries for uploaded_only reviews. 404 for unowned reviews.
Read-only. For outbound/hybrid v2 reviews, returns every paper the discoverer surfaced (across OpenAlex / arXiv / Exa), each with its initial relevance score, whether the fetcher acquired the PDF, and the screener's include/exclude verdict. Empty list for uploaded_only reviews. 404 for unowned reviews.
Read-only. Lists every Thoth systematic-literature-review run you own, with status, critic score, faithfulness score, and claim/citation counts.
Input schemas not visible in source code. Cannot verify parameter types, descriptions, constraints, or enums. Per HARD SCORING RULE, schema score must be 0 for all tools.
Output schemas not documented. Descriptions mention return values (e.g., 'returns the full markdown draft') but do not specify field names, types, or structure. LLMs cannot plan downstream tool calls without knowing response shape.
Descriptions lack WHEN/WHY context. E.g., 'Read-only. Lists every Thoth systematic-literature-review run you own...' does not explain when to call this vs get_review_draft or when to use pagination. Descriptions should guide tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
No error handling guidance visible. Descriptions mention '404 for unowned reviews' but do not explain what the LLM should do (retry, ask user, abort). Error responses must guide recovery.
Pagination not documented. list_reviews and list_discovered_papers likely return lists but no limit, offset, or cursor parameters visible. Large result sets risk context window exhaustion.