AG-UI

Open protocol that connects AI agents to frontend applications. Adopted by LangGraph, CrewAI, Mastra, Google ADK, and AWS Bedrock AgentCore. The wiki’s first reference to AG-UI; first appeared in the AIMock launch source as a flagship production user of aimock.

Stub — first appearance in the wiki via the AIMock launch. Will grow when a primary source covers AG-UI directly.

Why it’s in the wiki

AG-UI is the frontend-side counterpart to MCP (which connects agents to tools) and A2A (which connects agents to other agents). The three protocols together form the connective layer of the 2026 agentic stack:

ProtocolConnectsDirection
MCPAgents ↔ tools / data sourcesAgent reaches out to capabilities
A2AAgents ↔ other agentsAgent collaborates with peers
AG-UIAgents ↔ frontend applicationsAgent surfaces to a human user

The wiki should treat AG-UI as the third leg of this stool. As it gains adoption (already in 5 named frameworks), it becomes the standard way to ship agent UIs — important for any wiki coverage of agent frontend frameworks, CopilotKit specifically, and the broader 2026 “agentic UX” thread.

Production use

AG-UI’s end-to-end test suite uses AIMock with fixture-driven responses across LLM providers. The reference setup at apps/dojo/e2e/fixtures/openai is the canonical “AIMock at scale” example.

See Also