SandboxAQ integrates with Claude for drug design

SandboxAQ’s new Claude integration is less about inventing a new drug-design stack than about wrapping an existing one in a chat interface. On May 18, SandboxAQ said its Large Quantitative Models now connect to Anthropic’s Claude through the Model Context Protocol, or MCP, so users can run scientific workflows in natural language instead of writing custom code. The company said the setup is aimed at drug discovery, materials science and other quantitative R&D work. (prnewswire.com) The key claim is accessibility. SandboxAQ said running these models previously required specialized scientists and complex coding, while the Claude connection lets users move from a prompt to a simulation or workflow without managing infrastructure directly. Its documentation says the MCP server connects Claude and other compatible clients to SandboxAQ’s proprietary scientific models and is available to enterprise customers through onboarding and single sign-on setup. (prnewswire.com) What is actually being integrated? SandboxAQ is exposing its own quantitative models as callable tools inside Claude, not replacing core chemistry software with a language model. Claude acts as the interface layer; SandboxAQ’s models do the domain-specific computation. Anthropic’s connector page describes SandboxAQ’s offering as physics-based simulation combined with machine learning for scientific discovery, accessible through large language models. (claude.com) That matters because the drug-design workflow named in social posts — RDKit, PyMOL and AutoDock-style tooling — is made up of established components that already handle different parts of molecular work. RDKit is an open-source cheminformatics toolkit; its documentation includes a PyMOL interface for displaying and analyzing molecules. PyMOL, in turn, has long supported AutoDock-related plugins for setting up docking grids and viewing docking results. (github.com) In practice, that means a researcher could use Claude as an orchestration layer: generate or clean structures, inspect them visually, launch model calls, and track jobs through conversation rather than scripts. SandboxAQ’s Claude connector page shows example prompts for alloy optimization and job-status checks, while its MCP documentation says Claude can invoke the appropriate tool automatically once configured. The company’s current public documentation is clearest on AQCat Adsorption Spin and AQAffinity, which are SandboxAQ building blocks for catalyst screening and binding-affinity prediction. (claude.com) The part that needs caution is the broader social-media framing around “drug design with Claude.” SandboxAQ’s May 18 announcement explicitly highlighted drug discovery and materials science, but the most detailed public docs currently surfaced through its MCP pages emphasize AQCat and AQAffinity rather than a full public, end-to-end RDKit/PyMOL/AutoDock product page. So the verified development is a Claude-to-SandboxAQ model connection for scientific workflows; the fuller bench-scientist workflow described in community posts appears to be demonstrated through workshops, examples and surrounding toolchains rather than announced as a standalone commercial suite in the materials reviewed. (prnewswire.com) The next place to watch is SandboxAQ’s MCP documentation and Claude’s connector listing. SandboxAQ’s docs say additional scientific building blocks are part of the MCP server hub, and the Claude connector page is where Anthropic surfaces supported capabilities and example prompts for users who want to test the integration. (docs.aisim.sandboxaq.com)