MCP

The Model Context Protocol (MCP) is an open standard that lets AI applications connect to external tools and data through MCP servers, using one JSON-RPC message format instead of a custom integration for each system.

Anthropic open-sourced MCP on 25 November 2024 as a single protocol to replace one-off connectors between AI assistants and data sources (Anthropic, 25 November 2024). The architecture has three roles. An MCP host, the AI application, creates one MCP client for each MCP server it connects to. Servers expose tools the model can call, resources it can read and prompts it can reuse, and the messages follow JSON-RPC 2.0 (MCP documentation, accessed 9 November 2026).

The current protocol version is 2026-07-28 (MCP specification). MCP now sits in the Agentic AI Foundation, directed by the Linux Foundation, alongside the A2A protocol. In advertising, MCP has become the common way to expose buying and selling functions to agents. On 9 November 2026, 52 of the 53 entries in the IAB Tech Lab's agent registry were MCP servers (IAB Tech Lab registry, 9 November 2026).

Example: A buyer's agent connects to a seller's MCP server listed in the registry. It calls the get_products tool with a campaign brief, reviews the products returned, then calls create_media_buy with the chosen package and budget. Both tools appear in the server's public registry entry.

Related terms: A2A (Agent2Agent), agent registry, agentic advertising, AAMP, AdCP.

Authoritative source: Model Context Protocol, specification version 2026-07-28 (accessed 9 November 2026).

Read the full guide: What agent-ready ad infrastructure means.