MCP Tool

An MCP tool is a callable function exposed by an MCP server that an AI agent can invoke to perform actions in the world. Tools are the primary mechanism through which agents extend their capabilities beyond text generation — enabling them to interact with external systems, APIs, file systems, and services.

Anatomy of an MCP Tool

Each MCP tool has:

→ Name — A unique identifier (e.g., task, memory)
→ Description — Natural language explanation of what the tool does (used by the LLM to decide when to call it)
→ Input schema — JSON Schema defining the parameters the tool accepts
→ Response — The data returned to the agent after the tool executes

AgentRQ MCP Tools

AgentRQ exposes the following tools to Claude Code agents:

Tool	Actions	Replaces
`task`	create, get, updateStatus, reply, elicit, getAttachment	`createTask`, `getTask`, `updateTaskStatus`, `reply`, `elicit`, `getAttachment`
`memory`	load, save, delete	`loadMemory`, `saveMemory`, `deleteMemory`
`skills`	search, load, save, delete	`searchSkills`, `loadSkill`, `saveSkill`, `deleteSkill`
`workspace`	get, publishEvent	`getWorkspace`, `publishEvent`
`siteTools`	search, getDefinition, call	`listSiteTools`, `getSiteToolDefinition`, `callSiteTool`
`pano`	save, read, comment, attach, delete	—
`help`	—	—

Every tool takes two inputs, an action and a payload with that action's fields:

text

task {"action": "create", "payload": {"title": "Review the release notes", "body": "Check the notes before Friday.", "assignee": "human"}}

The help tool prints the exact payload: help {} lists the tools, help {"action": "task"} lists one tool's actions, and help {"action": "task.create"} shows one action's fields and an example. The old names in the Replaces column still answer for clients that are already installed.

How the Agent Uses Tools

When a tool is available, the LLM can choose to call it by generating a structured tool call in its response. The MCP client intercepts this, executes the tool call against the server, and returns the result to the model as a new context entry.

text

1. Model generates: task {"action": "reply", "payload": {"taskId": "...", "text": "Done!"}}
2. MCP client calls AgentRQ's task tool
3. AgentRQ delivers the message to the human's task board
4. MCP client returns: { "success": true } to the model
5. Model continues reasoning with updated context

Related Terms

→ MCP
→ MCP Server
→ MCP Client
→ Tool Use
→ Task