What is an MCP tool?

How an agent picks a tool

Each tool's description is effectively a prompt the model reads. A well-written description tells the agent exactly what the tool does and when to use it, so the model can match a user's natural-language request to the right tool without being told explicitly.

For example, the AetherWave server exposes aetherwave_generate_music, aetherwave_generate_image, and aetherwave_generate_video. When you ask for a song, the agent matches your intent to aetherwave_generate_music and calls it with the parameters it inferred.

Anatomy of a tool

A tool has three parts: a name (a stable identifier), a description (what it does and when to use it), and an input schema (the typed parameters it accepts). Many tools also carry safety annotations that tell the client whether the tool is read-only or makes changes.

The result a tool returns is structured so the model can reason about it, for instance a URL to a generated file, a status, and the credits spent.

Frequently Asked Questions

How many tools can a server have?

There is no hard limit. Servers commonly expose a handful to a few dozen. The AetherWave server exposes 16 tools across music, image, video, and audio.

Can I call an MCP tool directly?

Tools are designed to be called by the agent on your behalf. In most clients you simply describe what you want, though some clients let you invoke a tool explicitly.