What are MCP tools?

Tools are the things an MCP server lets your AI assistant do, like search, create or update.

MCP tools are the actions an MCP server offers to an AI assistant, such as searching a database, creating an issue or fetching a web page. The assistant decides when to call one, and the app should let you say no.

Also called MCP tool, MCP server tools, tools/list, tools/call, Model Context Protocol tools.

Each tool has a unique name, a short description for the model, and a list of inputs it accepts. Some tools also describe the shape of their output and carry optional annotations about their behavior.

When you ask for something, the assistant looks at the tools a connected server offers and picks one that fits. The app sends the call to the server and passes the result back. Tools are one of three things a server can offer. The other two are resources (read-only context) and prompts (ready-made instruction templates).

The spec says there should always be a human in the loop who can deny a call. In practice that is the approval prompt you see in Claude, ChatGPT or Cursor.

MCP tools are the actions an MCP server lets an AI assistant take, such as searching issues, creating a page or reading a file. The assistant chooses a tool from the list the server offers, fills in its inputs and asks the server to run it. The one thing to know first: a tool can change real data, so read the approval prompt before you accept.

Tools are model-controlled. The AI decides when to use them, based on your request. Resources, by contrast, are chosen by the app, and prompts are started by you.

A server that fails to do a tool's job reports an error in the tool result, so the assistant can see what went wrong and try again. The spec tells apps to show you a tool's inputs before the call and to ask for confirmation on sensitive actions. It also tells servers to check all inputs and limit how often tools can be called.

Good for

Anyone wondering what an MCP server actually does for them
Users deciding whether to approve a tool call
Developers designing tools for their own server
Admins reviewing what a connector can change

Comparison

Tools are the part of an MCP server that can change things, so they are where approval matters most. Resources only supply context.

Option	What it is	Who decides	Can change things
ToolsThis page	Actions	The AI model	Yes
Resources	Read-only context	The app	No
Prompts	Instruction templates	You	No

Questions

What are MCP tools?

The actions an MCP server offers to an AI assistant. Examples are searching a workspace, creating an issue or fetching a web page.

How does an assistant use an MCP tool?

It asks the server for its tool list, picks the one that fits your request and sends a call with the inputs. The server runs it and returns a result that the assistant reads and uses in its answer.

Who decides when a tool runs?

The AI model chooses, based on your request. The spec says an app should still let you deny a call, so many apps show an approval prompt.

What is the difference between tools, resources and prompts?

Tools are actions the AI can call. Resources are read-only data the app can pull in as context. Prompts are templates you start yourself.

Can an MCP tool change or delete my data?

Yes, if the server offers a tool that does. Tools can write to databases, call outside services and modify files, so read each approval prompt. Our advice: use a read-only option when the server page lists one.

Why does my assistant ask before using a tool?

The MCP spec says there should always be a human who can deny tool calls, and that apps should ask for confirmation on sensitive operations. Apps follow this with approval prompts.

How many tools can a server have?

A server can offer one tool or many, and the list can change while you are connected. Our Fetch page shows a one-tool server.

Where can I see real MCP tools?

Open any server page on this site and look at its tool list, or try the MCP Inspector to call tools yourself. See also our best MCP servers.

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

Tools (specification) · modelcontextprotocol.io
Understanding MCP servers · modelcontextprotocol.io