Anthropic Launches Claude Science – An AI Workbench for Scientific Research

On June 30, Anthropic announced Claude Science, a new AI workbench aimed at scientists that can run locally on macOS and Linux or on remote machines.

All users of Claude Pro, Max, Team, and Enterprise plans can access the tool (Team and Enterprise users require admin enablement), and Anthropic offers discounted Team‑level packages for research labs.

The platform’s stated goal is to unify the many databases and tools scientists switch between—such as PubMed, Jupyter, R, and command‑line terminals—by connecting Anthropic’s advanced model with those resources.

Claude Science does not introduce a new model; it is built on the standard Claude model. Users first interact with a general coordinating agent that can reach more than 60 databases and possesses a suite of related skills, and the tool can call any additional service via MCP.

Anthropic leverages Nvidia’s new BioNeMo Agent Toolkit to link to life‑science models and libraries, including Evo 2, Boltz‑2, and OpenFold 3.

“Professional scientific techniques remain owned by the partners that created them,” the company explained.

The system emphasizes reproducibility: every output includes an auditable generation history, and visualizations—such as 3D protein structures or genome‑browser tracks—are accompanied by the underlying code and full message logs.

Because Claude itself cannot execute large‑scale genomics or protein‑folding workflows, Claude Science can invoke SSH or Modal to interact with existing high‑performance‑computing clusters. The agent drafts a plan, the researcher reviews it, and Claude then submits the tasks to the lab’s resources, continuously monitoring results, flagging issues, and allowing session branching for alternative approaches.

In the broader AI‑for‑science landscape, Google unveiled Gemini for Science at its I/O conference, positioning it as a desktop science workbench that integrates over 30 life‑science databases and an “AI co‑scientist” engine. OpenAI previously offered Prism and GPT‑Rosalind for scientific writing and drug discovery, but after the departure of Kevin Weil in April, OpenAI discontinued the OpenAI for Science project and shifted focus to core services like Codex.