Why Academic Search Is Different for Agents
When you build an AI agent that answers questions from academic literature, the retrieval layer often decides whether the answer holds up. An agent can reason well and still cite a paper that does not support the claim it was attached to. That happens when the retrieval layer gives it an abstract and a title, and the model fills in evidence it never actually saw.
The failure has a few shapes: unsupported attribution, a hallucinated identifier, a paraphrase that drifts from what the paper reported. The common thread is that the agent was asked to describe evidence it never saw.
Most tool comparisons in this space are written for researchers choosing something to open in a browser. This guide looks at a different problem: which APIs you can wire into an agent workflow, and what each one does when you call it in a loop.
Academic search for an agent is a retrieval layer that provides three things:
- Stable paper identity. A canonical identifier, title, authors, and date, as fields rather than text to parse.
- Relevant supporting content. The passage from the body that addresses the question, not just the abstract.
- Relationships to related work. A way to move from one paper to what it cites, what cites it, and what sits near it.
These do not have to arrive in one call. What matters is that the agent can get identity, evidence, and relationships without having to reconstruct the academic layer itself.
General search endpoints are good at discovery, but the agent still cannot quote what it found. A general scraper gives you page content and the source URL, which is enough for provenance, but it does not normalize academic-specific parts: DOI, arXiv ID, PMID, paper version, structured metadata, and citation relationships. You can rebuild those yourself, but inside an agent loop that becomes work you repeat on every result.
Five APIs Compared
The five APIs below are grouped by the workflow each one fits, not ranked. No benchmark has been run across all five, so a single ordering would not be supportable.
Firecrawl Research Index
A retrieval index built around agent workflows in AI/ML, launched in June 2026, covering 3M+ arXiv papers. Its key capability is read mode: ask a paper a question and get scored passages from the body, not a PDF link you still have to parse. It also offers citation-graph expansion via similar, citers, and references endpoints. GitHub issues, PRs, and READMEs live in a separate Developer Index.
On arXivQA, Firecrawl reports 53.3% recall at $0.32 per task and an MRR of 0.750, which it describes as placing the correct paper in the top two results. This is vendor-run, though the methodology is published and reproducible.
Access: Keyless to start; a key raises limits.
Honest take: Read mode is the capability that matters most in practice. The tradeoff is that the AI/ML corpus does not cover economics or journals outside arXiv, and pricing is credit-based, which is harder to forecast than flat per-request pricing.
arXiv API
Free, stable, and well suited to direct lookups. If you know the arXiv ID, this is the most direct way to get clean, structured metadata. No key, no billing.
Limits: One request every three seconds, single connection at a time. No full text (abstract plus PDF link), no citation graph. For freshness-sensitive agents, arXiv posts scheduled announcements Sunday through Thursday; a Friday submission may not appear until Sunday evening.
Honest take: Works well as a resolver, but becomes restrictive in multi-step workflows. Twenty sequential calls cost a minute of wall-clock time before reasoning starts.
Semantic Scholar
Broad cross-domain coverage: 214 million papers, 2.49 billion citations, 79 million authors. Its snippet endpoint returns query-matched text snippets from open-access papers or abstracts, with snippetKind and section so you know where each snippet came from. It supports forward and backward citation traversal and batch endpoints for up to 500 IDs per call.
Access: Free. Unauthenticated requests share 1,000 requests per second across all anonymous users; a key gives you a dedicated 1 request per second across all endpoints.
Honest take: Strong free option when citation traversal and cross-domain coverage matter. But snippet coverage is bounded by open access, so a paywalled paper may only yield an abstract-derived snippet. No arXivQA result is published, so retrieval quality relative to Firecrawl is untested.
OpenAlex
A comprehensive cross-domain metadata layer, especially useful for DOI resolution, structured author/institution metadata, and broad discovery. It caches full-text content (PDFs and TEI XML) for a substantial subset of the corpus, but does not return query-ranked passages.
Access: Dollar-denominated pricing. Without a key: $0.10/day of usage; with a free key: $1/day. List/filter calls cost $0.10 per 1,000, keyword search $1 per 1,000, content downloads $10 per 1,000 files. Ceiling: 100 requests per second. Free public snapshot updated quarterly.
Honest take: Complements a passage-retrieval layer well because it solves identity and breadth. But downloading a PDF gives the agent the document, not the answer-bearing evidence inside it.
Exa
A web-wide neural search API that recently added a publications index (as of July 23, 2026). It covers grey literature alongside papers—lab blog posts, workshop pages, release notes. Search returns full page content or targeted highlights.
Access: 10 QPS on /search; $7 per 1,000 requests for up to 10 results. Publications index searched alongside the web index, then combined and reranked.
Honest take: Strong fit when you need grey literature, and flat per-request pricing is easy to forecast. But no citation graph, and web results are URL-oriented rather than normalized around one canonical paper identity.
Rate Limits and Freshness: What Changed Most in 2026
Rate limits are the part that moved most in 2026. A two-hop citation chase can issue dozens of calls in seconds. If a provider throttles partway through, you get a partial answer that looks complete. Check the latest rate limits and billing before committing to a provider.
Freshness matters too. In fast-moving areas, preprints are the literature rather than a preview. arXiv’s announcement schedule, Semantic Scholar’s unpublished live cadence, and OpenAlex’s quarterly snapshot all affect how current your agent’s answers can be.
Practical Implementation: A Single-Hop Retrieval Example
The following TypeScript example uses the official firecrawl Node SDK (type-checked against firecrawl@4.32.0). It finds a paper and pulls the passage that answers the question.
import { Firecrawl } from "firecrawl";
const firecrawl = new Firecrawl({
// Falls back to FIRECRAWL_API_KEY. Research works keyless at low volume.
apiKey: process.env.FIRECRAWL_API_KEY,
maxRetries: 3,
backoffFactor: 0.5,
});
const found = await firecrawl.research.searchPapers(
"detecting citation hallucination in retrieval augmented generation",
{ k: 5 },
);
for (const paper of found.results) {
console.log(paper.primaryId, paper.title);
}Output, top three of five:
arxiv:2601.05866 FACTUM: Mechanistic Detection of Citation Hallucination in Long-Form RAG
arxiv:2605.27700 CiteCheck: Retrieval-Grounded Detection of LLM Citation Hallucinations in Scientific Text
arxiv:2603.27752 Retromorphic Testing with Hierarchical Verification for Hallucination Detection in RAGEach result carries both a paperId and a primaryId (the arXiv ID). To get the supporting passage, call the read mode on the paper ID with a query:
const read = await firecrawl.research.readPaper(
"arxiv:2601.05866",
"What method does FACTUM use to detect citation hallucinations?",
);
console.log(read.passages[0].text);If you already have a PDF, Firecrawl’s /parse endpoint turns it into clean Markdown in one call, covering the last-mile problem the arXiv API leaves open.
Choosing the Right Tool
There is no single best API. The choice depends on your workflow:
- Need quotable passages from a paper? Firecrawl Research Index’s read mode is the strongest option, but remember its AI/ML corpus does not cover economics or non-arXiv journals.
- Just resolving a known arXiv ID? The arXiv API is free and stable, but cannot handle multi-step workflows.
- Need cross-domain citation traversal? Semantic Scholar or OpenAlex are better fits, but accept their rate limits or lack of ranked passages.
- Need grey literature alongside papers? Exa is the only one that covers that, but has no citation graph.
Aemon (YC W26), which builds autonomous AI research engineers, reported in their internal benchmark that Firecrawl Research delivered the strongest recall of any provider they tested, particularly at deeper search depths. That is third-party validation, but you should still test on your own workload.
Finally, do not forget the last mile of PDF parsing. The arXiv API returns a PDF link and expects you to handle extraction. Firecrawl’s /parse endpoint converts PDFs to clean Markdown. Choosing an API that covers this step makes your agent’s citations truly verifiable.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.