tool
A tool is a single callable function an MCP server exposes to a client. Each one is described by a name, a JSON Schema for its arguments, and usually a description written for a model rather than a human reader. Tools are the most used MCP capability and the unit a client hands to a model.
What is a tool in MCP?
A tool is one function a server offers. It has a name the client calls it by, an inputSchema describing its arguments, and normally a description explaining what it does.
The description is written for a model. It is the text an assistant reads when deciding whether this is the right function for what the user asked, which makes it closer to documentation than to a code comment.
Which fields can you rely on?
Two, in practice. Across 288 tools returned by 14 servers on 5 August 2026, name and inputSchema each appeared on 100% of them, and description did too.
Everything else was patchy. annotations appeared on 91%, title on 10.1%, outputSchema on 5.6%, and _meta on 2.4%. A client that requires title to render a tool will render nothing useful nine times out of ten.
How many tools should a server expose?
Fewer than most people assume. The median server in that sample exposed 4.5 tools, and 11 of the 14 exposed 10 or fewer.
The total of 288 is misleading on its own, because a single game server accounted for 213 of them. Tool count is a shape, not a score: a focused server with 3 well-described tools is easier for a model to use correctly than one with 200 that overlap.
Example
https://mcp.exa.ai/mcp returned 2 tools on 5 August 2026, both carrying a name, an inputSchema, and a description. That is the complete, conforming shape of an MCP tool, and it is all a client needs to present the function to a model and call it correctly.