A2A
Agent2Agent. An open protocol that lets software agents built by different companies find each other and work together.
What is A2A?
A2A, short for Agent2Agent, is an open communication protocol designed so that autonomous software agents from different vendors can discover one another, describe their skills and collaborate on a task. Rather than each pair of companies building a bespoke API connection, agents share a common format for capability descriptions, task requests, status updates and results. In advertising, A2A is one of the building blocks being explored for agentic workflows, where an agent representing a buyer negotiates with an agent representing a seller. It is often discussed alongside MCP, which connects an agent to tools and data rather than to other agents.
How A2A works
An agent publishes a short description of itself, called an agent card, at a known address. Other agents read it to learn what the agent can do and how to reach it, then exchange structured requests and replies.
- An agent publishes an agent card describing its identity, skills, endpoint and supported authentication.
- A client agent discovers the card and decides whether that agent can help with its task.
- The client agent sends a task request containing structured inputs, such as a campaign brief or inventory query.
- The remote agent processes the task, streaming progress or asking for more input if needed.
- The remote agent returns a result, such as available packages or proposed terms, which the client agent evaluates.
Why it matters
In advertising it gives a buyer’s agent a standard way to talk to a seller’s agent, without a custom integration for every pair of companies.
For advertisers
A shared protocol means a buying agent can reach many sellers without the advertiser funding a separate integration for each one.
For publishers
Exposing a standard agent interface makes inventory discoverable to automated buyers while the publisher still controls what its agent is allowed to offer.
For adtech platforms
Supporting a common protocol lowers integration cost and lets a platform participate in multi-vendor agent workflows.
Practical example
A buyer agent reads a publisher’s agent card, asks what video inventory is available next month and receives proposed terms.
Key considerations
- The protocol defines how agents talk, not what commercial terms are acceptable; business rules still have to be configured on each side.
- Authentication and permission scopes matter, because an agent may be able to commit budget or inventory on someone’s behalf.
- Adoption across advertising vendors is still emerging, so many workflows today combine agent protocols with existing programmatic pipes.
- Logging every request and response helps teams audit what an agent proposed and why.
More AI & Agentic Advertising terms
Frequently asked questions
What is an agent card?
It is a machine-readable description that an agent publishes so others can learn its name, skills, endpoint and how to authenticate before sending it work.
How is A2A different from MCP?
A2A connects one agent to another agent as a peer. MCP connects an agent to tools, data sources and systems it can call directly.
Does A2A replace OpenRTB?
No. OpenRTB handles real-time bidding on individual impressions, while agent protocols are being explored for planning and deal negotiation that happen before and around delivery.