Citation and retraction checker for AI agents. Give it DOIs, PubMed or arXiv ids and it says whether each paper exists and whether it was retracted, withdrawn or corrected, using Crossref and Retraction Watch. It also searches publisher articles and open research papers, shows the x402 price in USDC an agent pays to read or cite each one, and traces a passage to the page it was copied from. Free, read-only, no account.
Scored across 8 tools
Each tool has a stated, distinct role: discover browses a topic, search answers a question with verdicts, retrieve previews open-access papers, and lookup handles a single known identifier/URL. The only real friction is between search, retrieve and discover, which all return candidate article lists differentiated mainly by the presence of a verdict and the open-access restriction, so an agent may occasionally pick the wrong entry point.
All eight tools share a predictable naulon_ prefix followed by a lowercase snake_case verb or verb_noun phrase (appraise, discover, lookup, quote, retrieve, search, check_citations, trace_passage). The mix of bare verbs and verb_noun compounds is a minor stylistic deviation but remains readable and unambiguous.
Eight tools is well within a comfortable range and each maps to a distinct step in the workflow: find sources, appraise them, look up a known one, price it, verify citations, trace reuse. No tool appears redundant or like padding.
The surface covers discovery, appraisal, record lookup, price probing, open-access retrieval, citation/retraction checking and provenance tracing, which is a coherent lifecycle for a research-sourcing server. The deliberate absence of any purchase or full-text fetch tool is documented as by design, but it does leave the workflow ending at a price quote with no way to complete the transaction in-server.
Score how relevant each candidate is to the topic, from its free teaser alone — a 0..1 relevance plus a one-line rationale. Pass the candidates you got from naulon_discover (or a curated subset), and pass them back WHOLE — matchedInBody / matchedSemantic tell the scorer WHY the search returned each one, and dropping them makes a body-matched source with a terse teaser look identical to a near-miss with a keyword-ish one. FREE: it reads the teaser and those flags, never paid content, and it fetches and pays for nothing. Use it to decide what is worth quoting.
Inputs: topic (required), candidates (required). Returns: appraised
FREE citation and retraction check. For up to 25 references (a DOI, PMID, PMCID, arXiv id, or the reference as written), say whether each work exists and whether it was retracted, withdrawn, corrected or put under an expression of concern. Use it before you cite. Facts from Crossref and Retraction Watch; no judgement of the research. For a retracted or withdrawn citation, up to three reviews and trials that cite it are suggested; for an unfound one, up to three papers on its subject; each with its licence.
Inputs: references (required). Returns: _notice, results, provenAnswer
Browse a TOPIC: list candidate articles by title and teaser, with no verdict on whether any of them answers a question. Returns FREE public teasers only — slug, title, and summary — with no content and no payment. Call this first to see what is available before appraising or quoting. A candidate may carry WHY it matched: matchedInBody means your terms are inside the paid text but NOT in the title or teaser (strong — the teaser just does not show it), matchedSemantic means only that it is near your query in meaning and your terms are absent (weak: judge the teaser, and set aside one that is off-topic). NEITHER flag set means one of two opposite things, so read it against the rest of the response: if any OTHER candidate here carries a flag, this source searches, and no flag is the STRONGEST result of all — your terms matched the title or teaser, which is why no body-only or semantic explanation was needed. If NO candidate in the response carries a flag, the source does not search (a plain RSS feed), and the absence says nothing at all — judge every candidate on its teaser. Discovery returns the best available matches, never a promise that any of them answers your question.
Inputs: topic (required). Returns: note, refused, candidates
FREE. Get the record of one source you already have the address of: a publisher page by its URL, or a paper by DOI, PMID, PMCID or arXiv id. For a page it returns the title, site, summary and its read and citation prices. For a paper it returns the title, authors, venue, year, retraction and correction notices, licence, abstract where it may be shown, and where a free copy is. It searches nothing: to find sources by topic use naulon_discover or naulon_search.
Inputs: source (required). Returns: kind, found, paper, reason, +2 more
Probe the real x402 toll for a source WITHOUT paying — the free 402 price check. Returns the author price, the buyer's true total (when the publisher adds extra settlement legs such as an operator fee, the total is higher than the author price), and the settlement terms. gated:false means one thing only — the origin answered 2xx, so it is a genuinely free read you may just fetch. Anything else returns refused:true with the reason in note: a refusal to reach the url (off-gate identity, or operator policy such as a kill-switch or deny-list), a 404, an unreachable origin, or a malformed 402. A refused url is NEITHER payable nor free — do not fetch it and do not report it as free; retry or fix the url.
Inputs: slug (required), url. Returns: legs, note, asset, gated, +8 more
FREE preview, RESEARCH PAPERS only. For a question, lists the open CC0, CC BY, CC BY-SA or public-domain papers that could answer it and the price of each mode (answer, retrieve). This tool never pays and never runs the answer. Prices only: nothing on this server buys. How buying works: https://naulon.app/docs/for-buyers
Inputs: question (required), mode. Returns: how, why, fees, mode, +3 more
FREE. Ask a QUESTION: search every naulon publisher's articles and get each match priced, with its public summary, url, and a verdict on whether it answers the question and what reading it would add. For a plain list of a topic's articles with no verdict, use naulon_discover. Prices only: nothing on this server buys. How buying works: https://naulon.app/docs/for-buyers
Inputs: question (required), papers. Returns: next, quote, covers, papers, +10 more
FREE. Find which naulon publisher pages a passage was copied from, word for word. Returns each page's title, url, and whether it holds most of the passage or part of it, plus how many Citation Licenses the page has sold where its publisher shows that. It detects verbatim reuse only, not paraphrase or training. Send 40 words to 5000 characters.
Inputs: text (required). Returns: _notice, matches, searched
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Check AI references against Crossref and DataCite. DOI/arXiv metadata checks, plus tool discovery.
Catch AI-fabricated citations (real DOI + fake title). Retraction, open-access, 10,000+ CSL styles.
Hosted research subset: check citations, catch fake/retracted papers, search OpenAlex & PubMed.
Research papers for AI agents: PubMed, arXiv, Crossref; citations, references, open access texts
Add one secure layer between your agents and this server.
