MCP glossary
The vocabulary of the Model Context Protocol, defined for Amazon Ads practitioners.
- Model Context Protocol (MCP)
- An open standard that defines how AI applications connect to external tools and data sources through a uniform interface, so any compliant client can use any compliant server.
- MCP server
- A program that exposes a set of tools, resources, and prompts to MCP clients. An Amazon Ads MCP server exposes advertising operations (reporting, campaign edits) as callable tools.
- MCP client
- The AI application or agent (e.g. an assistant or IDE) that connects to MCP servers and invokes their tools on behalf of the user or model.
- Tool
- A discrete, callable action an MCP server exposes, for example "get_campaign_report" or "update_bid". Tools have typed inputs and outputs the model can reason about.
- Resource
- Read-only contextual data an MCP server can provide to the client, addressable by URI, such as an account structure or a saved report.
- Transport
- How a client and server communicate. MCP commonly uses stdio (local processes) or streamable HTTP / SSE (remote servers).
- Amazon Ads API
- Amazon’s official programmatic interface for managing advertising campaigns. MCP servers for Amazon Ads typically wrap this API and expose it as tools.
- Function calling
- A model capability for emitting structured calls to developer-defined functions. MCP standardizes and externalizes this so tools are reusable across clients rather than hard-coded per app.
- Managed MCP server
- A hosted MCP server run by a vendor, handling authentication, token rotation, and uptime so you don’t operate infrastructure yourself.