Read-only access to the ResearchThis Portal's curated findings feed and connected Zotero library via MCP protocol, proxying the deployed portal's HTTP API
Four read-only tools with clear verb-noun naming (search_*, get_*). All tools have descriptions (194 - 250 chars, within baseline 34 - 392 range). Input schemas present with type definitions and parameter descriptions. However, output schemas are not explicitly documented, responses are inferred from code (list[dict] or dict). Parameter descriptions are detailed but could be more concise. No error handling guidance or recovery paths documented. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.
Counts, freshness, and the theme-track taxonomy (id -> label/color) for the findings feed (GET /api/v1/meta on the deployed portal).
Findings already in Zotero, findings missing from Zotero, and DOI duplicates within Zotero -- an exact DOI/arXiv match between the two sources, computed by the deployed portal (app/reconcile_zotero.py), so "which findings are in both sources" or "what's missing from Zotero" can be answered directly instead of cross-referencing search_findings and search_zotero_library results by hand. Returns {"configured": False} if Zotero isn't configured on the deployed portal.
Search the curated findings feed (GET /api/v1/findings on the deployed portal). query: substring match against title or authors (case-insensitive). section: "relevant" or "related", or omit for both. track: a theme-track ID (see get_findings_meta for the taxonomy). warn: True to only include findings flagged unverified. linkedin: True to only include findings already used on LinkedIn. No-op if the deployed portal's feed manifest has linkedin_enabled off (see app/feed.py) -- the field is still present on every record regardless, just not filterable.
Search the connected Zotero library by title/creators substring (case-insensitive); an empty query lists the whole library. Returns an empty list if Zotero isn't configured on the deployed portal.
Output schemas not documented. Code returns list[dict] or dict, but field names, types, and structure are not specified in tool descriptions. LLMs cannot plan downstream operations or extract specific fields without trial-and-error.
No tool annotations. All four tools are read-only, but readOnlyHint is not set. This prevents clients from optimizing caching, retry logic, or safety checks.
No error handling guidance. _get_json() calls urllib.request.urlopen() without try-catch or error classification. If the portal is unreachable, returns raw exception. LLMs receive no recovery hints (e.g., 'Zotero not configured' for get_zotero_reconciliation).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Parameter descriptions are verbose (100+ chars for some). Baseline is 72 chars. Trim to essential constraints: 'substring match (case-insensitive)' instead of 'substring match against title or authors (case-insensitive).'
search_findings accepts optional boolean params (warn, linkedin) with default=None. Unclear if None means 'omit filter' or 'false'. Document: 'None = no filter; True/False = filter on/off.'