Search bioRxiv/medRxiv preprints, fetch metadata and revision history, resolve preprint→journal publication via @cyanheads/biorxiv-mcp-server

@cyanheads/biorxiv-mcp-server v0.1.15 — bioRxiv/medRxiv preprint search and metadata

Install: npm install @cyanheads/biorxiv-mcp-server Entry: node dist/index.js (stdio transport, ESM) Engine warning: requires Node ≥24 per package.json, but works fine on Node 22.

5 Tools

Key Gotchas

1. ⚠️ `cursor` must be a NUMBER, not string — passing "0" causes Zod validation error (expected number, received string). Use cursor: 0.
2. ⚠️ Two DOI prefix eras — older preprints use 10.1101/ prefix, newer ones (2025+) use 10.64898/. Both work with get_preprint and get_published_version.
3. ⚠️ Batch `get_preprint` with wrong `server` silently drops DOIs — if you specify server: "biorxiv" but include a medRxiv DOI, the medRxiv DOI silently returns no result (partial success, no error). Omit server to search both.
4. `search_preprints` uses EuropePMC under the hood — keyword relevance ranking comes from EuropePMC, then the server enriches matches with bioRxiv/medRxiv API data. This means search results include metadata like category and JATS XML that EuropePMC alone wouldn't have.
5. `search_preprints` with narrow date_from/date_to may return 0 results — the date filter is applied at the EuropePMC search level; if EuropePMC hasn't indexed recent preprints yet, results will be empty even though list_recent finds them.
6. `list_recent` pages 30 results at a time — response includes next cursor: N value; pass as numeric cursor param. total field tells you how many exist in the date range.
7. Category filter is case-sensitive — use exact capitalization from list_categories output (e.g. "Neuroscience" not "neuroscience").
8. `get_published_version` only works for preprints that have been published — unpublished preprints return MCP error -32001 with helpful recovery hint.
9. Output is Markdown-formatted — all tools return formatted Markdown with headers, bold fields, and structured layout.
10. Server logs JSON to stderr — verbose structured logging (pino) including per-tool latency metrics.

Verified Trace (14 calls, 100% success)

Batch 1 (5 calls):

• list_categories → 27 bioRxiv + 52 medRxiv categories (2ms)
• search_preprints "CRISPR gene editing" → 3 results, top: "Integrative Multiomic Analysis Reveals How Non-Viral Delivery System Selection Shapes CRISPR Gene Editing Outcomes" DOI 10.64898/2026.05.29.728905 (1127ms)
• search_preprints "COVID-19 vaccine efficacy" server=medrxiv date_from=2026-01-01 → 0 results (134ms) — EuropePMC date filter too narrow
• list_recent start=2026-06-20 end=2026-06-22 server=biorxiv cursor="0" → FAIL: Zod validation, cursor must be number (2ms)
• list_recent start=2026-06-20 end=2026-06-22 server=medrxiv → 23 preprints, top: "EpiLink: a simulation-based compatibility model" (633ms)

Batch 2 (9 calls):

• list_recent server=biorxiv cursor=0 (fixed) → 166 total, page of 30, next cursor: 30 (2158ms)
• list_recent category="Neuroscience" → 58 preprints in 3 days (1615ms)
• get_preprint DOI 10.64898/2026.05.29.728905 → full metadata + abstract + 1 revision (535ms)
• get_preprint batch [biorxiv DOI + medrxiv DOI] server="biorxiv" → partial