Skip to main content
Glama

Scholar Sidekick

by io.github.mlava
OverviewDiscussionsAdmin

Verify Citation

verifyCitation
Read-onlyIdempotent

Verify a claimed citation against the resolved record at its identifier. Detects the dominant AI-driven fabrication pattern documented by Topaz et al. (Lancet 2026): a real, resolvable identifier (DOI / PMID / PMCID / arXiv / etc.) paired with a title that does NOT correspond to the paper at that identifier. Use when the user pastes a citation and asks 'is this real?' or 'check this DOI' — most fabricated citations resolve cleanly under doi.org but their cited title and the resolved title disagree. Single citation per call. Required: title plus exactly one identifier (doi, pmid, pmcid, isbn, arxiv, issn, ads, or whoIrisUrl). Optional refinements: author (first-author family name), year, container (journal). Set screenWithLlm: true to invoke the Stage 3 LLM screen on low-confidence mismatches (catches informal-abbreviation false positives); LLM access is gated to authenticated first-party keys and paid RapidAPI tiers — anonymous callers get 400 LLM_SCREEN_FORBIDDEN. Returns: { verdict: 'matched' | 'mismatch' | 'not_found' | 'ambiguous', confidence: 'high' | 'medium' | 'low', matched: , mismatches: [{field, claimed, resolved, similarity}], candidates: [{item, registries, score}] (when title-search ran), provenance: {stages_run, resolved_via, registries_searched, llm_screen} }. Verdict semantics: 'matched' = claim agrees with resolved record; 'mismatch' = identifier resolves but title does not match (Topaz fabrication pattern); 'ambiguous' = identifier resolves to one paper but the claimed title matches a DIFFERENT paper found via title-search (CITADEL 'citation error' subtype — wrong identifier for a real paper); 'not_found' = neither the identifier nor the title resolves anywhere. No sibling tool overlaps: resolveIdentifier returns metadata for a known-good identifier; verifyCitation is the only tool that cross-checks claimed title vs resolved metadata. Read-only and idempotent — safe to retry. Works anonymously for the non-LLM path; the Stage 3 LLM screen requires authentication — set SCHOLAR_API_KEY (a free ssk key from https://scholar-sidekick.com/account) or use a paid RapidAPI tier. SCHOLAR_API_KEY also raises your rate limit.

Input Schema

NameRequiredDescriptionDefault
adsNoNASA ADS bibcode (19 chars).
doiNoDOI as cited (with or without https://doi.org/ prefix). Provide whichever identifier(s) the cited reference carries; the verifier uses the first one in priority order doi > pmid > pmcid > arxiv > ads > isbn > issn > whoIrisUrl.
isbnNoISBN (10- or 13-digit, hyphens tolerated).
issnNoISSN for journal-level resolution.
pmidNoPubMed ID as cited (digits only, or with 'PMID:' prefix).
yearNoPublication year as cited. Wrong year alone does not flip the verdict, but >=2-year gap from the resolved record lowers confidence.
arxivNoarXiv ID (e.g. '2301.08745' or 'arXiv:2301.08745'; old-style 'hep-ph/0501023' accepted).
pmcidNoPubMed Central ID (e.g. 'PMC1234567' or 'PMCID:1234567').
titleYesThe title as it appears in the cited reference. This is the field the verifier cross-checks against the resolved record at the supplied identifier. Required.
authorNoFirst-author family name as cited. Refines the verdict — a title-vs-resolved-title match plus an author mismatch raises suspicion of fabrication. Pass only the family name (e.g. 'Topaz', not 'Topaz, Maxim').
containerNoJournal or container name as cited (e.g. 'The Lancet', 'Neuroscience'). Soft signal — surfaced as a mismatch field but does not gate the verdict.
whoIrisUrlNoWHO IRIS URL (https://iris.who.int/...).
screenWithLlmNoOpt-in Stage 3 LLM screen. Fires only when the pre-LLM verdict is mismatch with low confidence (the informal-abbreviation false-positive bucket). Gated: requires an authenticated first-party API key or a paid RapidAPI tier; anonymous / free callers receive 400 LLM_SCREEN_FORBIDDEN. Default false.
semanticTitleNoOpt-in semantic title check (any tier). On a string-level matched verdict, a calibrated model judgment of whether the claimed and resolved titles describe the same work; catches one-word swaps in long titles that string similarity misses. A flagged title moves matched → ambiguous (review), never mismatch. Not byte-deterministic — _provenance.semantic_title carries the probability. Default false.

Output Schema

NameRequiredDescriptionDefault
matchedNoThe resolved record at the identifier, or null on not_found.
verdictNo
candidatesNo
confidenceNo
mismatchesNo
_provenanceNo

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changed
    • addedInput schema / properties / semanticTitle
      Added value: +{
+  "description": "Opt-in semantic title check (any tier). On a string-level matched verdict, a calibrated model judgment of whether the claimed and resolved titles describe the same work; catches one-word swaps in long titles that string similarity misses. A flagged title moves matched → ambiguous (review), never mismatch. Not byte-deterministic — _provenance.semantic_title carries the probability. Default false.",
+  "type": "boolean"
+}
  2. First observed

TDQS

A4.8/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already cover readOnly/idempotent/non-destructive/openWorld, yet the description adds substantial extra context: the LLM-screen auth gate and its 400 LLM_SCREEN_FORBIDDEN failure, the anonymous-vs-keyed rate-limit behavior, the required-exactly-one-identifier constraint, and full verdict semantics. This is well beyond the annotation baseline.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with purpose and trigger, then params, returns, verdict semantics, and auth. Nearly every sentence earns its place, though the returns block partially restates the existing output schema and the description runs long.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 14-parameter verification tool with an output schema, the description is complete: trigger, identifier selection, opt-in screens with their auth constraints, verdict taxonomy, and retry safety are all present. Nothing needed to invoke or interpret it correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is already 100%, so the baseline is 3, but the description adds real meaning: 'title plus exactly one identifier' with a documented priority order, plus when screenWithLlm and semanticTitle actually fire and what each can and cannot flip. It does not add new per-parameter syntax beyond the schema, so it stops short of a 5.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a precise verb+resource ('Verify a claimed citation against the resolved record at its identifier') and immediately names the distinguishing failure mode it targets. It explicitly differentiates itself from resolveIdentifier, so an agent can route without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives concrete trigger conditions ('when the user pastes a citation and asks is this real? or check this DOI') and an explicit no-overlap statement contrasting with resolveIdentifier. It also bounds scope with 'Single citation per call'.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

All Scholar Sidekick toolsInspect
Try in Browser
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources