jwm0203/claude-science-for-win — explained in plain English
Analysis updated 2026-05-18
Run a Claude Science style research workflow on Windows instead of macOS or Linux.
Query public science databases like PubMed, UniProt, and ChEMBL through one command line tool.
Get an independent reviewer agent to check citations and verify figures against code before finalizing a research output.
Trace any generated figure or number back to exactly which command and input data produced it via the run manifest.
| jwm0203/claude-science-for-win | abhisumatk/epstein_files_rag | asdfo123/forgewm | |
|---|---|---|---|
| Stars | 34 | 34 | 34 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | hard |
| Complexity | 4/5 | 3/5 | 5/5 |
| Audience | researcher | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
No license is stated, some bioinformatics command line tools have limited Windows support and may need WSL2 or a remote Linux machine.
Claude Science for win is an attempt to recreate Claude Science, a research workbench Anthropic released for macOS and Linux only, so that it also works on Windows. The README is upfront that Claude Science itself is not a new model, but an orchestration layer combining a coordinator agent, over 60 research skills and connectors, an independent reviewer step, and audit trails, and this project rebuilds that same orchestration layer for Windows using research skill packs, database connectors, and an auditable history system. The one thing it cannot reproduce is the official graphical interface, since this version is used through the command line and Claude Code instead. The architecture mirrors the original: a coordinator agent takes a research request and plans and delegates work, six specialist agents handle areas like literature review, single cell analysis, CRISPR screening, phylogenetics, protein structure, and cheminformatics, and a separate reviewer agent independently checks citations, recalculates numbers, and verifies that figures match the underlying code before anything is considered finished. Every analysis run gets its own folder recording start and end times, every tool call made along the way, input and output files registered with checksums, and a snapshot of the software environment, so results can be traced back to exactly how they were produced. It ships working connectors for public databases including PubMed, Europe PMC, arXiv, UniProt, RCSB PDB, Ensembl, Reactome, ClinVar, and ChEMBL, which the README says have been tested and return real data. Connectors that need a paid account or GPU access, such as Benchling, CellGuide, or BioNeMo's structure prediction models, are included as placeholders that report a clear message about what access they still need rather than pretending to work. The project runs on Python through two optional conda environments, one for general research tasks and one for biology specific tools, and the README notes that a few command line bioinformatics tools have limited Windows support and may need to be run through WSL2 or a remote Linux machine instead. No license is stated in the README.
A Windows-native reimplementation of Anthropic's Claude Science research workbench, rebuilding its coordinator, specialist, and reviewer agent architecture with auditable run tracking and public science database connectors.
Mainly Python. The stack also includes Python, Conda, Claude Code.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.