Papers with Code for the lab sciences · early access

The code behind lab science, found and verified for AI agents.

The script that reads a Thermo raw file, fits a qPCR curve or cleans up Raman spectra is sitting in a research repo nobody can find. Bench Beacon maps lab-science code to papers, instruments, file formats and deposited data, and checks that every link is real.

Join the waitlist Read llms.txt

Machine learning had Papers with Code. The lab sciences never did.

Thousands of labs publish working code for their instruments, assays and papers. Agents can search GitHub; they can't tell which repo handles their format, whether it matches the paper, or whether the code is even there.

No map by method

Code search knows languages and stars, not assay, instrument or file format. Curated registries cover polished tools and miss the long tail of lab code.

Claims that don't hold

Plenty of repos promise "code to reproduce the paper" and don't contain it. Nobody checks at scale.

Code and data drift apart

The data lands in GEO, PRIDE or Zenodo. Which code made which figure from which accession is recorded only in the repo, if anywhere.

Search is easy. Verification is the product.

Every record says where each field came from, so your agent knows what to rely on and what to check.

Code-present verdictsWhen a paper or README promises code, we check whether it's actually there: yes, partial or no.
License verdictsyes, conditional or unknown, with the reason, plus a caveat when a software license is applied to data.
Provenance on every fieldEach inferred value is tagged rule, llm or pending, with a confidence where the model gives one. Nothing is guessed.
Paper and data linksPaper ↔ repository ↔ GEO/SRA/PRIDE/Zenodo accession. Start from a DOI or a file format and follow the chain.
Pointers, not copiesWe return file locations and header hints, never file contents. Your agent fetches from the source and honors the license.

// find_tools(format=".abf")
{
  "id": "gh:lab/patch-clamp-toolkit",
  "target_method": {"v": "ABF patch-clamp files",
                    "src": "llm", "conf": 0.92},
  "handles_format": {"v": "yes", "src": "rule",
                     "why": "reader + .abf test fixtures"},
  "paper": {"doi": "10.xxxx/example",
            "code_present": "yes"},
  "license": {"spdx": "MIT", "reusable": "yes"},
  "maintained": {"v": "active", "src": "rule"},
  "readiness": 88
}

Illustrative record, not a real repository. Verdicts are informational, not legal advice.

Read-only tools, over MCP and REST.

Compact JSON built for agents. Plug it into Claude, a custom agent or a plain HTTP client.

`find_tools`	Give a file format, instrument or assay; get the repos that verifiably handle it, ranked by readiness.
`find_by_paper`	Give a DOI, arXiv ID or bioRxiv DOI; get its repos, whether the promised code is there, and the linked data accessions.
`check_paper_code`	A reproducibility verdict for one paper: what was claimed, what exists, what's missing.
`search`	Free text plus assay, instrument, format, discipline, license and paper filters, with the reason each result matched.
`get_repo_record`	The full record: every inferred field with its source, readiness breakdown, paper and data links, file pointers.

Built for owners, too

Repositories flagged for secrets or personal data are excluded. Owners can ask us to correct or remove a record at any time.

Made by a working scientist

Bench Beacon is built by a multi-omics postdoc who has lived this problem, not by a crawler that treats every repo the same.

Free for exploring

A free daily allowance for researchers, and prepaid credits for teams whose agents run all day.

Get early access.

We're opening to a small group of agent builders and research teams first. Tell us which file format or instrument gives you trouble and we'll prioritize it.

Prefer email? Write to support@benchbeacon.org.