bioRxiv

What the bioRxiv connector does

The bioRxiv connector puts the bioRxiv and medRxiv preprint servers inside Claude, ChatGPT and any MCP-compatible agent. Its seven read-only tools find preprints by date and subject, pull full metadata by DOI, track which preprints later appeared in journals, filter by funder, and report platform statistics. It needs no sign-in.

bioRxiv tools (7)

• search_preprintsRead-only

  Search bioRxiv/medRxiv preprints. Returns DOI, title, authors, abstract preview, category. WHEN TO USE: - Literature review and research discovery - Finding recent research in specific fields - Tracking new submissions by date or category SEARCH METHODS (use only ONE): - Date range: date_from + date_to (e.g., '2024-01-01' to '2024-06-30') - RECOMMENDED - Recent days: recent_days=30 (last 30 days) - Recent count: recent_count=50 (searches last ~90 days, returns up to 'limit' results) IMPORTANT: All searches use date ranges internally. recent_count searches a 90-day window. If no search method specified, defaults to last 60 days. SERVERS: - 'biorxiv': Biological sciences (default) - 'medrxiv': Medical/health sciences CATEGORIES (27 available): biochemistry, bioinformatics, cancer biology, cell biology, genetics, genomics, immunology, microbiology, molecular biology, neuroscience, and 17 more. Use get_categories tool for full list. LIMITATIONS: - NO keyword/text search - filter by category and date only - Results are NOT peer-reviewed preprints EXAMPLES: - Last 30 days of neuroscience: recent_days=30, category='neuroscience' - Cancer biology Q1 2024: date_from='2024-01-01', date_to='2024-03-31', category='cancer biology' PAGINATION: Use cursor for next page (cursor=100 for results 100-199) RELATED: get_preprint (full details by DOI), get_categories (list all categories)

• get_preprintRead-only

  Get complete metadata for a specific preprint by DOI. WHEN TO USE: - You have a DOI and need full details - Need abstract, authors, PDF URL, funding info - Checking if preprint was published in a journal RETURNS: title, all authors, corresponding author + institution, full abstract, category, license, version, PDF URL, web URL, funding details, published DOI (if available) DOI FORMATS (all accepted): - '10.1101/2024.01.15.123456' - 'https://doi.org/10.1101/2024.01.15.123456' SERVERS: - 'biorxiv': For bioRxiv preprints (DOI contains biorxiv dates like 2024.01.15) - 'medrxiv': For medRxiv preprints IMPORTANT: Preprints are NOT peer-reviewed RELATED: search_preprints (find DOIs), search_published_articles (find journal version)

• get_categoriesRead-only

  List all 27 bioRxiv subject categories for filtering searches. WHEN TO USE: - Before search_preprints to see valid category values - To understand research area classifications RETURNS: Category names and API-compatible format (e.g., 'cancer biology' -> 'cancer_biology') FULL LIST: animal behavior and cognition, biochemistry, bioengineering, bioinformatics, biophysics, cancer biology, cell biology, clinical trials, developmental biology, ecology, epidemiology, evolutionary biology, genetics, genomics, immunology, microbiology, molecular biology, neuroscience, paleontology, pathology, pharmacology and toxicology, physiology, plant biology, scientific communication and education, synthetic biology, systems biology, zoology

• search_published_preprintsRead-only

  Find preprints that have been published in peer-reviewed journals. WHEN TO USE: - Track which preprints became journal articles - Find the peer-reviewed version of a preprint - Analyze preprint-to-publication patterns - Filter articles by whether they were published by a specific journal SEARCH METHODS (use only ONE): - Date range: date_from + date_to - Recent count: recent_count=50 - Recent days: recent_days=30 FILTERS: - include_details: True for full metadata (slower), False for summary (faster) - publisher: Filter by DOI prefix (e.g., '10.1038' for Nature) - see below COMMON PUBLISHER PREFIXES: - '10.1038' - Nature Publishing Group - '10.1126' - Science/AAAS - '10.1016' - Elsevier - '10.1371' - PLOS - '10.7554' - eLife - '10.1073' - PNAS EXAMPLE: Find recently published medRxiv papers: server='medrxiv', recent_days=30

• search_by_funderRead-only

  Search for preprints funded by a specific organization using ROR ID. WHEN TO USE: - Track research output from specific funding bodies - Analyze funder-specific publication patterns - Monitor grant-funded research areas REQUIRES: funder_ror_id (9-character ROR ID) COMMON FUNDER ROR IDS: - '021nxhr62' - NIH (National Institutes of Health) - '01cwqze88' - NSF (National Science Foundation) - '02mhbdp94' - European Commission - '029chgv08' - Wellcome Trust - '05a28rw58' - HHMI (Howard Hughes Medical Institute) - '006wxqw41' - MRC (Medical Research Council UK) - '00f54p054' - BBSRC (UK) - '01s5ya894' - Chan Zuckerberg Initiative Find more ROR IDs at: https://ror.org/search RETURNS: DOI, title, authors, abstract preview, category (same as search_preprints) IMPORTANT: Date range required. Earliest date is 2025-04-10 (funder metadata inception) SERVERS: 'biorxiv' or 'medrxiv'

• get_content_statisticsRead-only

  Get bioRxiv submission statistics over time. WHEN TO USE: - Analyze bioRxiv growth trends - Track submission patterns (new vs revised papers) - Generate platform statistics reports RETURNS (per month/year): new_papers, revised_papers, new_authors, cumulative_papers, cumulative_authors, period (YYYY-MM or YYYY) INTERVALS: 'monthly' (default) or 'yearly' NOTE: Returns all historical data from bioRxiv inception to present

• get_usage_statisticsRead-only

  Get bioRxiv usage/engagement statistics over time. WHEN TO USE: - Analyze readership trends - Track engagement metrics (views, downloads) - Compare abstract vs full-text vs PDF engagement RETURNS (per month/year): abstract_views, full_text_views, pdf_downloads, cumulative_abstract, cumulative_full_text, cumulative_pdf, period (YYYY-MM or YYYY) INTERVALS: 'monthly' (default) or 'yearly' NOTE: Returns all historical data from bioRxiv inception to present

Read from the server on 2026-08-18, including each tool's own safety annotations.

Limits

• Unrefereed is the default state of every result, and this is the sharpest limit here. bioRxiv's screening covers plagiarism and non-scientific or dangerous content, not scientific validity. An agent that presents a preprint finding as established produces confidently wrong output, and in a clinical or public-health context that is a real harm. Anthropic operating the connector adds nothing here, and neither do we.
• Withdrawals are nearly invisible. No tool has a withdrawal or retraction field. Only get_preprint carries the optional type value that can read withdrawn, and search results omit it. Because posted manuscripts cannot be removed, withdrawn work stays searchable indefinitely.
• There is no keyword, author or full-text search. search_preprints filters by date and category only, and says so in its own description. "Find preprints about X" has to become a category plus a date window plus your own filtering.
• The category filter does not work for medRxiv. The 27-value enum is bioRxiv's subject list, and openRxiv's API silently ignores a mismatched category rather than erroring — returning unfiltered results. Search medRxiv by date and server only.
• The statistics tools are bioRxiv-only. Neither takes a server parameter, and openRxiv's content-summary endpoint returns bioRxiv figures regardless of any server argument. Its usage endpoint serves medRxiv separately, but no tool reaches it.
• Funder search cannot see anything before 2025-04-10, the inception of funder metadata in openRxiv's submission system. Earlier grant-funded work is unreachable by funder however it was funded.
• The connector returns metadata and abstracts, not full text. get_preprint returns PDF and web URLs; no tool retrieves or parses the manuscript body.
• Publication linkage is not a quality verdict. A missing published_doi may mean under review, or simply unrecorded. Of 43,632 bioRxiv preprints posted in 2024, 18,960 had a linked journal publication when we checked.
• A DOI is not a stable handle on content. Preprints are revised in place under new version numbers and the DOI resolves to the latest — which is why all six corpus tools are openWorldHint: true. Record the version alongside anything you cache or cite.
• Documented DOI examples use a superseded prefix. Tool descriptions show 10.1101/...; new postings use 10.64898/.... Both are live. Never construct a DOI from the pattern.
• The vendor's own tool lists are stale in four places — directory, tutorial, instructions and one output schema. The live list is the only reliable one.
• The Claude Code plugin points at a third-party endpoint (mcp.deepsense.ai) rather than the directory's Anthropic-operated host. That host returned a CDN 403 to our probe, so we establish nothing about it either way. This page describes the directory endpoint only.
• No tool writes anything, and none takes personal data. All seven are read-only, and no input schema accepts a manuscript, comment, credential, account or patient identifier. The connector cannot post a preprint, flag one, or correct the record.
• We did not call any tool. Our check was a read-only initialize, tools/list and prompts/list handshake; every statement about tool behaviour comes from the server's own descriptions and schemas. Corpus counts come from openRxiv's public API directly.

Frequently asked questions

Sources

• Live initialize, tools/list and prompts/list handshake against https://hcls.mcp.claude.com/biorxiv/mcp, anonymous, 2026-08-18. serverInfo reported biorxiv 3.2.0. Source for the tool count, every tool description and parameter named here, the uniform read-only/non-destructive/idempotent annotations, the six-versus-one openWorldHint split, the absence of any withdrawal or retraction field, the category and server enums, the search_by_funder date floor, the instructions block naming funder_search and get_statistics, the outputSchema naming search_published_articles, and the analyze_preprint prompt and its seven-point rubric. resources/list returned JSON-RPC -32601 "Method not found", consistent with the server declaring prompts and not resources. No tool was called. · retrieved 2026-08-18
• Endpoint auth discovery, 2026-08-18 — /.well-known/oauth-protected-resource at the host root, the RFC 9728 path-suffixed form, the same path under the connector prefix, and /.well-known/mcp/server-card.json all returned 404, consistent with the anonymous handshake and the directory's authless record. · retrieved 2026-08-18
• Cross-server prompt and resource probe — docs/marketing/data/mcp-prompts-probe.json, 30 readable directory servers, probed 2026-08-18. Source for six servers serving live prompts, fifteen serving resources, and four — bioRxiv, ChEMBL, Exa and NPI Registry — advertising zero while serving one to three. · retrieved 2026-08-18
• openRxiv API documentation (retrieved 2026-08-18; 200). Source for the /details, /pubs, /pub, /publisher, /funder, /sum and /usage endpoints the connector fronts, /pub being bioRxiv-only, the 2025-04-10 funder-metadata inception date, and the ROR ID format. Confirms the three phantom directory names map onto upstream endpoints. · retrieved 2026-08-18
• openRxiv API queried directly rather than through the connector, all 2026-08-18. /sum/y returned 345,705 cumulative bioRxiv preprints, 130,011 revisions and 43,632 new preprints in 2024; /pubs/biorxiv/2024-01-01/2024-12-31/0 totalled 18,960; /details/medrxiv/ for 2024 reported 12,863 new medRxiv papers against /pubs/medrxiv/'s 4,590. /details/biorxiv/10.1101/2020.01.30.927871/na/json returned two versions typed new results and withdrawn. A medRxiv sample returned clinical categories absent from the tools' enum, and the same query with a bioRxiv category returned category: all — the filter is ignored, not rejected. /details/biorxiv/30d returned 255 records all prefixed 10.64898 against a uniformly 10.1101 June 2024 window. Source for every corpus figure here. · retrieved 2026-08-18
• bioRxiv FAQ (retrieved 2026-08-18; 200). openRxiv's own statement, and the source for everything this page says about peer review, screening and withdrawal: preprints not certified by peer review, edited or typeset; the warning that they may contain errors and report information not yet accepted or endorsed; the multi-step screen and the not-certified statement and PDF stamp; the policy that posted manuscripts cannot be removed, that withdrawal posts a statement and watermarks every PDF version, and that removal has occurred only in extremely rare legal or safety cases; bioRxiv and medRxiv being co-managed with separate scope and screening policies and dual posting not permitted; the refusal of manuscripts with identifying images; and bioRxiv being operated by openRxiv, founded by Cold Spring Harbor Laboratory and funded by the Chan Zuckerberg Initiative, CSHL, the Sergey Brin Family Foundation and a university consortium. www.biorxiv.org/robots.txt carries no Content-Signal line and its User-agent: * block allows content paths at crawl-delay 7, disallowing only Drupal system paths; it separately lists ClaudeBot, GPTBot and CCBot under Disallow: /. Those are training-crawler directives rather than a rule on a single directed read, and the facts above are restated in our own words with attribution. No preprint text or abstract is reproduced anywhere on this page. · retrieved 2026-08-18
• Anthropic's bioRxiv connector tutorial (retrieved 2026-08-18; 200), the documentation URL the directory publishes. Source for the install paths, the worked prompts behind the four use-case sections, and the statement that preprints are screened for plagiarism and non-scientific content but are NOT peer-reviewed — and, as stale material, for "nine primary tool calls", the three names the server does not expose, the "over 310,000 preprints" figure and the openRxiv attribution. claude.com/robots.txt is Allow: / with no Content-Signal line. · retrieved 2026-08-18
• Life sciences plugin marketplace — and (retrieved 2026-08-18; 200 each). Source for the manifest naming https://mcp.deepsense.ai/biorxiv/mcp and deepsense.ai as author against the directory's Anthropic endpoint, and for the entry carrying no deprecation notice. The repo's biorxiv directory holds only .claude-plugin — no vendor-authored skill exists for this connector. · retrieved 2026-08-18
• deepsense.ai endpoint probe — https://mcp.deepsense.ai/biorxiv/mcp returned an HTTP 403 CloudFront page rejecting the request method (2026-08-18). Nothing was established about what that host serves, and no claim is made about it. · retrieved 2026-08-18
• Anthropic connector verification and directory documentation — and (retrieved 2026-08-18; 200 each). Source for verification not being a security audit, and for the label reflecting the level of review a connector received rather than how it works. · retrieved 2026-08-18
• Anthropic privacy policy — , the privacy_policy the directory publishes here. Support is .
• Anthropic Connectors Directory entry (added 2026-01-07; snapshot 2026-08-16). Lists eight tool names, an empty prompt_names array, anthropic tier, life-sciences, is_authless: true, Read permissions, streamable-http and Anthropic as author. Source for the directory side of the drift, the absent advertised prompt, and the host grouping drawn from the same 1,897-entry snapshot. · retrieved 2026-01-07