SandboxAQ brings its drug discovery models to Claude — no PhD in computing required

A scientist at a drug company wants to know how a molecule will behave inside a human cell. A few years ago, answering that question meant either running a wet lab experiment for weeks or handing the problem to a specialist who could write code to run a simulation. Most small biotech teams had neither the time nor the person. SandboxAQ thinks that gap is now closable with a conversation in Claude.

What happened

SandboxAQ, a science-focused AI company that spun out of Alphabet in 2022, has connected its suite of drug discovery tools to Claude, Anthropic's AI assistant. The integration means researchers can now run SandboxAQ's molecular simulation and analysis models by typing plain-language requests into Claude, rather than writing code or learning a specialized interface.

The tools in question focus on a field called molecular modeling. This is the process of predicting how drug compounds will interact with proteins in the body. Getting that prediction right early in the research process can save years of lab work. Traditionally, running these models required either deep computing knowledge or a dedicated bioinformatics team (a group of specialists who combine biology and data science).

SandboxAQ's bet is that the bottleneck in drug discovery is not the quality of the models. Other companies, including Chai Discovery and Isomorphic Labs, have been racing to build better and better prediction models. SandboxAQ is focused on a different problem: most of the scientists who need these tools cannot actually use them without technical help.

The Claude integration works through something called MCP, or Model Context Protocol (a standardized way for AI assistants to connect to external tools and data sources). When a researcher asks Claude a question about a molecule, Claude can reach out to SandboxAQ's models in the background, run the relevant simulation, and return the results in plain language. The researcher does not need to know anything about the underlying system.

SandboxAQ announced the integration in May 2026. The TechCrunch report covering the news notes that the company is positioning this as an access story rather than a performance story. The models are not necessarily newer or more accurate than what competitors offer. The argument is that they are now reachable by people who previously could not reach them.

Why it matters

Drug discovery is a useful lens here, but the bigger idea applies well beyond pharma.

Think about any field where powerful analytical tools exist but most practitioners cannot use them without a specialist in the room. That describes a lot of science, a lot of medicine, and honestly a lot of business analysis too. The pattern is familiar: a great model sits inside a system that requires technical training to operate, so only a small group of people ever benefit from it.

What SandboxAQ is doing is separating the model from the interface. The model still runs on their infrastructure. But the interface is now a conversation. A medicinal chemist who has spent their career in a lab, not at a keyboard, can describe what they are trying to understand and get a useful answer back.

For anyone building AI-powered products today, this is a concrete example of a design choice worth paying attention to. The technical capability is table stakes. The question is whether a non-specialist can actually use it without a tutorial, a manual, or a help ticket.

The MCP approach that makes this possible is also worth understanding at a basic level. Anthropic published the Model Context Protocol as an open standard last year, and a growing number of tools are now connecting to Claude and other AI assistants through it. What this means practically: the same conversation interface you might already use for writing or research can, if the right tools are connected, also run specialized software in the background. The assistant becomes a kind of universal front door.

For small biotech startups, academic labs, or even hospital research teams that cannot afford a dedicated data science hire, this kind of access shift is real. A tool that previously required a specialist to operate is now operable by the person who actually has the scientific question. That changes what a small team can attempt on a given Tuesday.

The broader pattern, companies competing on access rather than on raw model performance, is likely to show up in other technical fields over the next year or two. Legal research, financial modeling, materials science. Wherever there is a gap between who has the analytical tools and who has the domain expertise, a plain-language interface starts to look like a solution.

What to do

If you work in any field that uses specialized analysis software (biology, chemistry, engineering, finance), spend 20 minutes this week asking Claude what tools it currently has access to through MCP connections. You can simply type: "What external tools or data sources can you connect to right now?" The answer will depend on how your Claude account is set up, but it gives you a map of what is already possible.

If you are building a product that sits on top of a complex or technical tool, look at Anthropic's MCP documentation to understand how other developers are wrapping specialized capabilities in conversational interfaces. The SandboxAQ approach, keep the powerful model, replace the complicated interface, is a template that travels well.