Skip to content
// Glossary · Protocol and platform

MCP tools

Also called: tool calls, MCP capabilities, tool discovery

What are MCP tools?

MCP tools are the named capabilities an MCP server exposes to a connected agent. Each one has a name, a description, and a typed input schema, so an agent can discover what is available at connect time and decide during a session which to call. The agent needs no integration code written for that specific server.

// In more depth

What a tool actually is.

Three parts, and the third is the one that makes discovery work.

01

A name and a description

Written for the model rather than for a developer. The description is how an agent decides whether this tool is the right one for what it is trying to do, so vague descriptions produce unused tools.

02

A typed input schema

Declares the arguments and their types. The agent fills them in from context, and the schema is what stops it inventing a call shape the server cannot answer.

03

A result the model reads

The return value goes back into the session as something the agent reasons over, not as a payload for application code to parse. That shapes how results should be written.

// Why it matters

Why this design matters.

Tool discovery is the difference between an integration per tool and an integration per protocol.

No per-client work

A server that speaks MCP works with every compatible client. Nobody writes a Cursor version and a Claude Code version of the same capability.

Capabilities can change

Because tools are discovered at connect time rather than compiled in, adding one does not require every client to update.

Calls stay inspectable

A tool call is a discrete, named event. That is what makes it something a permission gate can pause on and an audit log can record.

// Commonly confused with

Commonly confused with.

Three things that are not MCP tools.

An API endpointAn endpoint is called by code that was written to call it. A tool is chosen by a model at runtime from a description, which is why the description matters as much as the schema.
A pluginPlugins are usually specific to one host application. Tools belong to a server that any compatible client can connect to.
Function callingClosely related, and it is the underlying model capability. MCP is the protocol that standardises how tools are advertised and reached across clients.
// FAQ

MCP tool questions

How does an agent know which tools exist?

It asks at connect time. The server returns its tool list with names, descriptions, and input schemas, and the agent decides from there which are relevant to the task in front of it.

Do I write different tools for different coding agents?

No, and that is the point of the protocol. One MCP server is reachable from every compatible client, so Claude Code, Cursor, Codex, and others all see the same tools.

What tools does Virex expose?

Virex Memory exposes memory tools so an agent can write and retrieve organization memory. Virex Remote exposes a session and decision surface so an orchestrating agent can follow a run and act on it.

Are tool calls recorded?

In Virex Remote, tool-call activity is part of what the session view shows and what the transcript keeps, so a run is reviewable afterwards rather than only watchable live.

Does calling a tool cost extra with Virex?

No. Virex bills a flat $15 per seat per month per product with no usage component.

Tools your agents can already reach.

Virex Memory is an MCP server, so connecting a compatible client is the whole integration.