MCP Advertising

MCP advertising is the practice of serving advertising to or through agents that communicate via the Model Context Protocol (MCP) specification.

MCP advertising is the practice of serving advertising to or through agents that communicate via the Model Context Protocol (MCP) specification.

What it means.

The Model Context Protocol is an open specification that standardizes how AI agents connect to external tools, data sources, and services. It was introduced by Anthropic in late 2024 and has been adopted across a wide range of agent hosts and tool providers, with over 4,000 public MCP servers listed by early 2026.

MCP advertising refers to ad delivery that moves through this protocol. In practice, an MCP server exposes a tool the agent can call — for example, get_surface — that returns a sponsored recommendation matched to the current query and context. The agent then renders that recommendation as a disclosed placement inside its output.

MCP itself is not an advertising protocol. It is general-purpose plumbing. Advertising behaviors are layered on top by specific server implementations and by conventions the industry is still defining. The disclosure and manipulation rules that make advertising honest are not yet written into the MCP specification. They have to be enforced by the server operators and by the agent hosts that install those servers.

How it differs from advertising for AI agents generally.

Advertising for AI agents is the broad category. MCP advertising is one transport inside that category — the one where the agent-to-ad-server communication rides on MCP rather than on a proprietary SDK, a direct REST call, or another protocol. Other transports remain common in 2026, including vendor-specific SDKs and webhook integrations.

The reason to distinguish MCP specifically is interoperability. An MCP-based ad integration can be plugged into any MCP-compatible agent host without custom work. The same server that serves ads to Claude Desktop can serve them to Cursor, to a community-built agent, or to any future host that speaks MCP. That portability does not exist for SDK-based integrations, each of which binds the ad system to a specific host.

Why it matters.

MCP advertising is emerging. The protocol is new and widely adopted, but the commercial conventions around it — bid formats, disclosure requirements, measurement standards — are still being defined. Decisions made in the next 12 to 24 months will shape the category for years.

Surfacedd is tracking MCP advertising closely. We have not yet shipped an MCP product. When we do, we will only serve disclosed Surfaces through MCP. That means every placement will carry a visible label, every placement will be structurally separate from the agent's organic output, and the publisher (the agent developer who installs the MCP server) will be able to remove the system at any time without penalty.

The risk to watch is manipulation at the protocol layer. Tool responses that contain instructions to the model — rather than just data — can nudge an agent's organic output toward a sponsor without the user seeing anything labeled. A disciplined MCP advertising standard prohibits that pattern explicitly. Until such a standard exists, honest operators self-enforce.

For the longer discussion of what good MCP advertising looks like and the patterns we will not ship, see MCP Advertising, Explained.

Related terms.

1. Advertising for AI Agents
2. AI Ad Network
3. AI-Native Advertising