Skip to main content
Back

LAP Fills the Missing Link Between AI Agents and Lab Instruments

LAP proposes a missing third protocol layer that lets AI agents safely command real laboratory instruments, filling the gap left by MCP and A2A.

The agent protocol stack has two well-known layers: Anthropic's Model Context Protocol (MCP) handles how an agent connects to tools and data, and Google's Agent2Agent (A2A) handles how agents talk to each other. But there is a third edge that neither protocol touches: the connection between an AI agent and a physical laboratory instrument. A new paper from Lin Zhu and colleagues proposes the Lab Agent Protocol (LAP) to fill that gap.

The missing edge of the agentic triangle

Autonomous science is moving from demonstration to infrastructure. Large language model agents now plan experiments, and self-driving laboratories execute them. But the infrastructure connecting those agents to the instruments they command has never been standardized. Every such system rebuilds the link between the reasoning agent and the physical instrument from scratch, against fragmented vendor SDKs and standards built for deterministic software clients rather than probabilistic, goal-directed agents.

The agent protocol ecosystem already has two of its three edges covered. MCP handles how an agent connects to tools and data sources. A2A handles how agents communicate with each other across organizational and platform boundaries. Neither models the agent-to-instrument edge, where operations are stateful, safety-critical, exclusively owned, physically embodied, and produce measurements with units, calibration, and uncertainty.

The reason existing protocols fall short is not a minor oversight. Physical instrument operations consume real reagents and irreplaceable samples; they take minutes to hours; they can injure people or destroy equipment; they require exclusive access while running; and they produce results that are only meaningful when accompanied by physical units, a calibration reference, and an uncertainty. None of these properties is expressible in an MCP tool schema or an A2A AgentCard.

What about existing lab standards?