AI agent protocol glossary
The vocabulary of AI agent protocols, defined in plain language. Terms are grouped by category; many link to a fuller guide.
Concepts
A software system that uses a model (often an LLM) to pursue goals with some autonomy — planning, calling tools, and acting over multiple steps rather than answering a single prompt.
A shared set of rules for how agents exchange messages, tasks, tools or context. Also the name of a specific community REST specification (Agent Protocol).
The pattern of an agent invoking external functions, APIs or data sources during a task. MCP standardises how those tools are described and called.
The ability for agents, tools and platforms from different vendors to work together through shared protocols instead of one-off custom integrations.
How one system learns what another can do at runtime — for example, listing MCP tools or reading an A2A agent card before delegating a task.
Coordinating multiple steps, tools or agents toward a goal — deciding what runs, in what order, and how results flow between components.
An HTTP interface that lets clients start, monitor and retrieve results from an agent — either a bespoke API or one following a spec like Agent Protocol.
Protocols
An open protocol standardising how agents and LLM apps connect to external tools, data sources and context, using JSON-RPC and a client–server model.
An open protocol for communication and task delegation between independent agents, using agent cards to advertise capabilities and tasks to coordinate work.
A framework-agnostic REST API spec for driving autonomous agents through runs, tasks, steps and artifacts — an API-style interface to an agent.
A2A
A machine-readable document (in A2A) describing an agent's identity, capabilities and endpoints, so other agents can discover and call it.
Foundations
A lightweight remote-procedure-call protocol encoded in JSON. Both MCP and A2A build on JSON-RPC-style request/response messages.
An architectural style for HTTP APIs based on resources and standard methods (GET, POST…). The Agent Protocol spec and many AI agent APIs are RESTful.
A model capability where the LLM emits a structured call to a named function with arguments. It underpins tool use and maps naturally onto MCP tools.