ToxMCP is the open-source toxicology tool layer used by NGRA.ai, a next-generation risk assessment platform. It connects AI assistants to chemical-safety tools and returns structured, source-linked results with visible uncertainty and screening limitations.
Each module uses the Model Context Protocol (MCP), an open standard that lets compatible AI assistants call external tools. You can use a module directly, combine several modules in a workflow, or review the code and evidence without using an AI assistant.
- ToxMCP suite guide — understand the modules and choose a starting point
- Computational Toxicology (CompTox) tools — try source-linked chemical identity and screening evidence from the U.S. Environmental Protection Agency
- Preprint — read the scientific background
- Chemical identity and screening evidence: CompTox MCP
- Exposure scenarios: Direct-Use Exposure MCP
- Environmental fate: Environmental Fate MCP
- Kinetics and internal dose: Physiologically Based Pharmacokinetic (PBPK) MCP
- Mechanistic pathways: Adverse Outcome Pathway (AOP) MCP
- Chemical grouping and read-across: Organisation for Economic Co-operation and Development (OECD) QSAR Toolbox MCP, where QSAR means quantitative structure-activity relationship
- Rapid property screening: ADMETlab MCP, covering absorption, distribution, metabolism, excretion, and toxicity
- source links and provenance that can be reviewed by a person
- explicit assumptions, limitations, and uncertainty
- structured results that can move between compatible tools
- clear separation between screening evidence and stronger scientific conclusions
These tools support research and screening. A runnable model or returned result is not automatically scientifically qualified, clinically appropriate, or ready for a regulatory decision. Review important outputs independently and follow each upstream data provider's access, rate-limit, license, and attribution terms.
Contributions are welcome. Start with the repository that owns the tool you want to change and read its local guidance. Organization-wide defaults are available in our contributing guide and security policy.
ToxMCP was developed in part through the VHP4Safety project and related computational-toxicology research. Funding included the Dutch Research Council (NWO), grant NWA.1292.19.272.