# MCP tool

Canonical page: /glossary/mcp-tool

API keys & secrets glossary, Claude Code & MCP. MCP Tool: What Claude Is Actually Calling

**Short answer:** An MCP tool is a single named action that an MCP server offers, complete with a description and a schema for its inputs. When Claude Code decides a tool is relevant, it calls it with specific arguments and gets a result back, the same way it would call any other function.

## Why it matters

A server can expose several tools. Claude Keychain's server exposes list_secret_names, which returns names only, and run_with_secrets, which runs a command you specify with a named secret injected as an environment variable and returns just the command's output.

## In Claude Keychain

Claude Keychain's tool descriptions explicitly warn against running commands that dump the environment, such as env or printenv, because that is the one way a secret could still end up in Claude's context.

## Common questions

**Can Claude call a tool without asking me?**

Claude Code normally asks for permission before running a tool that can change something or access data, and you can configure how strict that is.

**Can a tool see previous tool calls?**

No. Each call gets only the arguments it is given and returns only its own result.

## Related

[MCP server](/glossary/mcp-server), [ANTHROPIC_API_KEY](/glossary/anthropic-api-key), [Subprocess](/glossary/subprocess)