dsh-biorxiv

bioRxiv preprint search, version/publication status and export tools for DeepSeek Harness (dsh). Zero runtime dependencies.

Status: 0.1.0-alpha — functional core, not a 1.0 release. Relevance evaluation is provisional; do not treat it as peer-reviewed evidence.

Tools

• biorxiv_search — search bioRxiv preprints by topic, optional date window and field. Default minScore=3 filters low-confidence text matches; set minScore=0 for higher recall.
• biorxiv_paper — DOI lookup: latest version, publication DOI/journal, retraction flag when available.
• biorxiv_latest — latest preprints by date window and optional category.
• biorxiv_export — export search results or one DOI as JSON, Markdown, CSV or BibTeX.
• biorxiv_digest — structured markdown digest (extractive by default; optional opt-in Chinese summaries; no automatic delivery).

All tools return a stable envelope:

{
  "ok": true,
  "data": {},
  "error": { "code": "", "message": "", "retryable": false, "hint": "" },
  "meta": { "requestId": "...", "elapsedMs": 12, "cached": false, "source": ["europepmc"], "warnings": [] }
}

On failure ok is false, data is {}, and error.code is one of E-001–E-010.

Install (one command)

dsh plugin --profile headless add github:TYEclipse/dsh-biorxiv#v0.1.0-alpha.20

Replace headless with the target profile. dsh plugin add is a pnpm forwarder: a --profile is required, and pnpm must be on PATH.

Build from source instead:

pnpm install
pnpm build
pnpm pack
dsh plugin --profile headless add "file:$PWD/$(ls -t dsh-biorxiv-*.tgz | head -1)"

Verify:

dsh plugin --profile headless list
dsh --profile headless "Use biorxiv_search to search for 'protein language model' and report the first title and DOI."

dsh plugin add is a pnpm forwarder: a --profile is required, and pnpm must be on PATH.

CLI

dsh-biorxiv search "protein language model" --limit 5
dsh-biorxiv paper 10.64898/2026.07.21.739963
dsh-biorxiv latest --days 2 --limit 10
dsh-biorxiv export "hippocampal ripple" --format markdown --out refs.md
dsh-biorxiv digest "protein language model" --limit 5

Each command prints one JSON envelope and exits 0 on success, 1 on a tool error, 2 on a usage error.

MCP

Run the stdio server:

dsh-biorxiv-mcp

MCP client config example:

{
  "mcpServers": {
    "biorxiv": {
      "command": "npx",
      "args": ["-y", "--package", "dsh-biorxiv", "dsh-biorxiv-mcp"]
    }
  }
}

An npm package is not published yet. The npx form above works after publication; for a GitHub install, point the client at node ~/.dsh/profiles/<profile>/node_modules/dsh-biorxiv/dist/apps/mcp/main.js.

The server exposes biorxiv_search, biorxiv_paper, biorxiv_latest, biorxiv_export and biorxiv_digest with the same JSON envelope.

Configuration

Defaults live in the plugin schema; no config is required.

{
  "cachePath": "~/.cache/dsh-biorxiv/cache.db",
  "cacheMode": "auto",
  "timeoutMs": 10000,
  "retries": 2,
  "maxBytes": 2000000,
  "userAgent": "dsh-biorxiv/0.1 (DeepSeek Harness plugin)",
  "summaryEnabled": false,
  "summaryModel": "deepseek-chat",
  "summaryBaseUrl": "https://api.deepseek.com",
  "summaryApiKeyEnv": "DEEPSEEK_API_KEY"
}

• summaryEnabled: default false. When true, digest --summary sends only public title/abstract to the configured model and labels output as machine-generated.

• The API key is read from the named environment variable; it is never written to config, cache or logs.

• CLI: dsh-biorxiv digest "topic" --summary opts in; it also enables the provider for that process.

• MCP: set DSH_BIORXIV_SUMMARY=1 in the server environment, then call biorxiv_digest with summary: true.

• cacheMode: auto (SQLite when available, JSON fallback), sqlite, or json.

• Cache keys are SHA-256 hashes; query text is not written into the cache key.

• Delete the cache file to reset; it is disposable.

Data sources and boundaries

• Search: Europe PMC REST (SRC:PPR + PUBLISHER:"bioRxiv").
• Latest/version details: bioRxiv public API.
• Journal/retraction enrichment: Crossref and OpenAlex; failures are non-blocking.
• No full-text PDF download, no scraping behind 403/paywalls, no writes to source sites.
• Do not use for clinical, medical or investment decisions. Preprints are not peer-reviewed.

Development

pnpm test
pnpm coverage

The coverage ratchet is enforced by scripts/coverage-ratchet.mjs; line coverage must not fall below .coverage-baseline.json.

License

MIT. See LICENSE. Privacy and data flow: see PRIVACY.md.