Anthropic debuts Claude Science, a specialized AI workbench for life sciences research

Anthropic has unveiled Claude Science, a dedicated application designed to streamline how researchers conduct scientific work. Available in beta for macOS and Linux, the platform operates either locally or on remote machines, functioning as a unified workspace where scientists can integrate their existing tools and databases without constant context-switching.

A consolidated research environment#

Scientists typically juggle multiple platforms—PubMed, Jupyter notebooks, statistical software like R, terminal commands, and specialized databases—throughout a single project. Claude Science aims to eliminate this fragmentation by serving as a connective layer that brings these disparate resources together. The platform leverages Anthropic's standard Claude models paired with a coordination agent that can access over 60 life sciences databases and integrate with additional services via Anthropic's Model Context Protocol (MCP).

Rather than building new scientific models from scratch, Anthropic relies on established partnerships. The system incorporates tools from Nvidia's BioNeMo Agent Toolkit to access specialized life sciences models—including Evo 2, Boltz-2, and OpenFold3—while keeping domain expertise where it belongs. As the company notes, "The specialized science stays with the partners who built it."

How it works#

Claude Science functions as a reasoning and coordination layer. When researchers begin working, they interact with a generalist agent that understands how to navigate available tools and databases. The platform excels at handling repetitive analytical work: literature review, multi-step research execution, and manuscript refinement.

All outputs include an auditable creation history, allowing researchers to validate and reproduce results. Users can chat with the agent about specific details, annotating figures and manuscripts inline. The underlying code for any generated visualizations—including 3D protein structures and genome browser tracks—remains accessible and transparent.

Computational power and integration#

Claude Science cannot natively handle large-scale genomics pipelines or protein-folding computations, so it integrates with existing high-performance computing infrastructure. Researchers can connect their SSH clusters or Modal accounts; the agent drafts a computational plan for human review, then submits jobs to the lab's available resources. An accompanying agent monitors execution and flags issues, and sessions can be forked at any point to test alternative approaches.