SAP Agent Gateway Explained: The New Runtime for SAP AI Agents

When SAP announced the Business AI Platform at Sapphire 2026, one component generated more technical questions than any other: the SAP Agent Gateway. It is the runtime layer that sits between AI agents and your SAP backend systems — and it changes how integration teams should think about building SAP AI automation.

Here is a plain-English breakdown of what it does, how it relates to MCP, and what state it is actually in as of mid-2026.

What the Agent Gateway actually is

The SAP Agent Gateway is a runtime service on SAP Business AI Platform that handles two types of communication:

• →Agent-to-SAP: An AI agent calls a SAP backend capability — read a purchase order, post a goods receipt, check a cost centre balance. The Gateway translates the agent's request into the appropriate SAP call (RFC, OData, BAPI) and returns a structured response.
• →Agent-to-Agent (A2A): One AI agent hands off a task to another agent — a procurement agent escalating to an approval agent, for example. The Gateway manages the handoff using the A2A protocol.

Before the Agent Gateway, every team built their own wiring: custom middleware calling RFCs, OData services wrapped in Python scripts, one-off integrations that worked until something changed. The Gateway is SAP's attempt to standardise this at the platform level.

A2A: the protocol underneath

The Agent Gateway uses A2A (Agent2Agent), an open protocol developed collaboratively by Google, SAP, and other enterprise vendors and now stewarded by the Linux Foundation. A2A defines how agents advertise their capabilities, accept tasks, and return results.

Think of A2A as the HTTP of agent communication — a common language that any compliant agent can speak, regardless of which AI model or framework it is built on. SAP has committed fully to A2A as its preferred inter-agent standard. This means that agents you build today using A2A-compliant frameworks will work with SAP's ecosystem without proprietary lock-in.

How MCP and A2A work together

MCP and A2A are often mentioned in the same breath, and it is worth being precise about what each one does:

In a typical SAP AI workflow: an orchestrator agent receives a user request via A2A, calls your SAP system via MCP to retrieve data, processes it with an LLM, and returns the result. MCP handles the SAP data layer; A2A handles the agent coordination layer.

Current status and what is not yet GA

As of May 2026, the SAP Agent Gateway is live but with important limitations:

• —Outbound A2A (SAP agents calling external agents) is currently available
• —Inbound bidirectional A2A (external agents calling SAP via the Gateway) is expected GA Q4 2026
• —Joule Studio 2.0 integration with the Gateway is in controlled rollout
• —SAP Domain Models, which would make agent responses significantly more SAP-aware, are expected GA Q3 2026

For teams building SAP AI automation right now, this means the architecture is clear and implementable — but certain capabilities require designing with Q4 2026 GA dates in mind. Projects that start architecture work now will be in the best position to use the full bidirectional Gateway when it ships.

What to do with this information

If you are planning SAP AI automation in 2026, the practical implications are:

• ✓Build your SAP data access layer using MCP — it is stable, model-agnostic, and the Gateway is designed to work with it
• ✓Design multi-agent workflows with A2A in mind even if full Gateway GA is months away — retrofitting is expensive
• ✓Do not build direct RFC/OData wrappers for new projects; the Gateway pattern will supersede them
• ✓Factor SAP Domain Models (Q3 2026) into your timeline — they will significantly reduce the prompt engineering burden for standard SAP processes