Documentation
Nullary is a structured, citable record of what didn't work in drug discovery — inactive compounds, terminated trials, failed screens, and developability dead ends — delivered through an MCP server and a REST API.
Why negative results
Failed experiments rarely get published, and when they do they're scattered across incompatible databases. That absence is expensive: teams re-run dead ends and models train only on successes. Nullary normalizes negative findings across public sources into one schema, keyed to compounds and targets, with provenance on every record.
Two ways in
- MCP server — install into Claude, Cursor, or any MCP client and query in natural language. Install the MCP server →
- REST API — call
api.nullary.aidirectly from pipelines and apps. REST API reference →
Start here
- Quickstart — your first query in two minutes.
- Data model — modalities, finding types, outcomes, and provenance fields.
- Live demo — search the dataset in your browser, no install.