Documentation

Open Agent Protocol — a standard interaction surface so any caller can send commands to any service, and any observer can consume the events it produces.

Caller

Any Caller

app · agent · IoT · human

Command

→

OAP Endpoint

Your Implementation

/.well-known/oap

Event

→

Consumer

Any Agent

LLM · app · UI · agent

Expose /.well-known/oap

Serve a discovery manifest describing your agents, capabilities, and transport bindings. Consumers read it once and know everything.

Accept commands, produce events

Callers POST commands to your service. Your service processes them (any way it likes) and the events it produces are observable via the REST or MCP API.

Observe every execution

Every command→event cycle is stored as an execution trace — input, output, duration, and success — giving you a full audit trail.

Getting Started

Protocol OverviewWhat OAP is and how it works
DiscoveryThe /.well-known/oap manifest

Agents

Global RegistryFuture concept — public OAP directory
LifecyclePause and resume agents
EventsEvents published by services
CommandsAccept and catalogue commands
QueriesSynchronous reads of current state
MemoryView agent memory state

Transports

RESTHTTP/JSON API reference
MCPModel Context Protocol binding
A2AAgent-to-Agent protocol binding

Reference

VersioningHow versions work
ConformanceWhat it means to be OAP-compliant
Design DecisionsWhy OAP is shaped the way it is

Comparisons

OAP vs UCPHow OAP relates to Google's Universal Commerce Protocol