Capability Discovery

Capability Discovery

Also called: 能力发现 · Capability negotiation

An agent querying at runtime what capabilities a server or peer agent currently offers, instead of hard-coding the capability list in code.

ConnectQuerycapabilitiesUse what'savailableno hard-coding
Schematic (simplified)

How it works

After the connection is established, the two sides negotiate: the client asks "which features / tools do you support," and the server returns a current list (in MCP, tools/list and resources/list; in A2A, reading the Agent Card). The actions the model then sees are this runtime list, not a build-time assumption.

Why it matters

It enables dynamic composition: a server adds a tool and the client can use it on the next connect with no client code change; different environments (local / production) can expose different capabilities and still adapt. The cost is an extra handshake and having to handle "the peer's capabilities differ from what I expected."

Common misconceptions

  • Capability discovery is not one-time configuration — it is runtime negotiation on each connection.
  • Discovered capabilities still need authorization and validation — "can be listed" does not mean "free to call."

Related terms

Sources

  1. Model Context Protocol — Architecture: capability negotiation
  2. A2A Project — Agent Card / discovery

Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.