Glossary
What is MCP (Model Context Protocol)?
MCP is an open protocol, originated by Anthropic, that standardises how AI applications and agents connect to external tools, data sources and other agents — the substrate that agent-to-agent advertising standards build on.
The problem it solves
A model on its own can only reason over what is in its context. To do anything useful in the world it needs tools — to query a database, call an API, read a document, complete a transaction.
Before MCP, every one of those connections was bespoke. Each application defined its own way of describing what a tool did and how to call it, so a tool built for one assistant had to be rebuilt for the next. The integration cost scaled with the product of tools and applications.
MCP standardises the interface. A tool exposed over MCP can be used by any MCP-speaking client, which turns that multiplication into addition.
What it standardises
- Discovery. How a client finds out what a server can do.
- Invocation. How a tool is called and how arguments are described.
- Resources. How data sources are exposed for a model to read.
- Results and errors, so a client can handle failure consistently.
- Transport, so the same server works over different connection mechanisms.
Why advertising standards build on it
AdCP needs agents on the buy and sell side to discover each other, exchange structured proposals, and transact. That is exactly the problem MCP already solves, and re-solving it inside an advertising specification would have produced a protocol nobody outside advertising implements.
Building on MCP means an advertising agent is an ordinary MCP participant. The same client that can query a database or read a document can negotiate a media buy, and the advertising-specific part is limited to the vocabulary rather than the plumbing.
What it means for a brand
The practical implication is about being callable. If your product catalogue, availability and pricing are exposed over MCP, an agent completing a task on a user's behalf can use them directly — which is the mechanism behind an agentic conversion.
If they are not, the agent falls back to whatever it can scrape or recall, which is where stale prices and invented policies come from. Being machine-callable is becoming an availability question rather than a technical nicety.
Common questions
Is MCP only for Anthropic's models?
No. It originated at Anthropic and is an open protocol with implementations across the ecosystem, including clients and servers built by other vendors.
Do I need MCP to advertise inside AI assistants?
Not today — buying platforms handle the integration. It matters if you want your own systems callable by agents, or if you are building on the sell side.
More in ai and conversational advertising
How buying, selling and measurement work when the surface is a conversation.
- AdCP (Advertising Context Protocol)
- Agent-to-Agent Attribution (A2A attribution)
- Agentic advertising
- Agentic conversion
- AI ad network
- AI DSP (AI demand-side platform)
- AI media
- AI open web
- AI publisher
- AI share of voice (AI SOV)
- AI SSP (AI supply-side platform)
- AI walled gardens
- Answer insertion
- Artificial intelligence (AI)
- Branded agent
- ChatGPT Ads
- Citation bidding
- Conversational ads
- Conversational frequency cap
- GEO / AEO (Generative engine optimization / Answer engine optimization)
- Hallucination liability
- Intent density
- Machine learning
- Post-click ad economy
- Prompt inventory
- Sponsored follow-up
- Sponsored intelligence (SI)
- Sponsored prompt
- Synthetic audience
- Trust graph