Labwire: An Open Protocol for AI-Controlled Lab Instruments

Labwire is an open, MCP-inspired protocol that lets AI agents discover a laboratory instrument's capabilities, issue commands, stream telemetry, and walk away with a cryptographically signed record of what happened. Now at v0.3 draft, it defines instrument commands as JSON Schema so any agent framework can drive hardware without bespoke integration code.

Safety and physical typing are built into the wire protocol itself: every quantity carries mandatory UCUM units, actions are classified S0-S3 by reversibility and hazard level, irreversible or hazardous commands require operator confirmation or a bound grant, and errors are typed with retryability. v0.3 adds typed resources and references, letting agents discover physical state, like a liquid handler's deck layout, and act correctly without extra prompt engineering.

The reference implementation ships three realistic simulated instruments, a Python SDK, an MCP adapter, a CLI for verifying ed25519-signed run bundles, and bridges for ophyd (Bluesky) and PyLabRobot devices. A closed-loop demo shows a scripted or Claude-driven optimizer tuning experiment parameters across simulated hardware and producing verifiable signed evidence of the winning run.

For engineers building self-driving labs, Labwire offers a vendor-neutral, auditable alternative to today's bespoke instrument integrations, with safety and provenance treated as first-class protocol features rather than add-ons.