SandboxAQ Routes Quantum Chemistry Simulations Through Claude

On May 18, 2026, SandboxAQ launched a Claude integration for its Large Quantitative Models (LQMs). This architecture leverages the Model Context Protocol (MCP) to let Anthropic’s models orchestrate physics-based simulations. Researchers can now execute drug discovery and materials science computations using natural language, bypassing the traditional requirement for specialized coding skills and custom infrastructure.

Physics-Grounded Simulation Models

The system relies on SandboxAQ’s proprietary LQMs, which train on high-fidelity simulation data rather than human text. This training pipeline incorporates quantum chemistry, molecular dynamics, and microkinetics, utilizing tools like OpenFold and computing infrastructure developed alongside NVIDIA.

The integration launches with AQCat Adsorption Spin, a model designed for catalyst discovery. It calculates adsorption energy to determine how strongly molecules bind to a catalyst surface. This metric dictates the viability of reactions necessary for sustainable aviation fuel, green hydrogen production, and plastics recycling.

SandboxAQ will soon deploy additional pharma-focused models to the Claude environment. AQPotency screens thousands of compounds to prioritize drug candidates based on efficacy. AQCell maps pathway activation and performs toxicity screening.

Orchestration and Accessibility

Claude functions as the reasoning engine and orchestrator for these external models. When a user requests a specific chemical simulation, Claude interprets the plain-English parameters, formats the required API calls via MCP, and executes the LQM payload. Nadia Harhen, GM of AI Simulation at SandboxAQ, notes this approach compresses computational setup from weeks down to hours.

This architecture removes the infrastructure friction that typically bottlenecks AI inference in specialized scientific domains. SandboxAQ intends to capture non-specialist researchers and executives across biopharma, energy, and advanced materials. The company estimates this broader “quantitative economy” represents a $50 trillion sector.

Market Positioning

SandboxAQ operates with significant capitalization, having raised over $950 million since spinning out of Alphabet five years ago under chairman Eric Schmidt. The Claude integration reveals a distinct commercial strategy compared to competitors.

While firms like Chai Discovery and Google DeepMind’s Isomorphic Labs compete primarily on base model performance, SandboxAQ is prioritizing user interface and platform distribution. By embedding its proprietary simulations within Claude’s established chat interface, the company bypasses the need to build and maintain a standalone scientific software platform. Partha P. Mukherjee from Purdue University confirms this strategy successfully bridges the gap between scientific intuition and rigorous computation.

If you manage scientific compute pipelines, evaluate how MCP integrations change your infrastructure requirements. Shifting orchestration to a hosted LLM allows your team to focus computational resources on running the proprietary LQMs rather than maintaining the connective middleware.