Built-in Agent Tools (Apprentice MCP)

Every Apprentice agent ships with a built-in tool surface. These tools are exposed to the agent over Apprentice's own MCP layer, so the agent calls them the same way it calls any MCP tool. You do not install them. They are part of the runtime.

This page explains what the built-in tools are, how they are turned on or off, and where they run. The family pages that follow list every tool by name so you can design a custom agent and know exactly what it can do.

Built-in Tools Versus External MCP

Apprentice has two tool sources:

Built-in tools: shipped with the app and grouped into families such as memory, tasks, messaging, files, and browser. Covered on these pages.
External MCP tools: provided by MCP servers you install and bind to an agent. Covered in Understand MCP Tools in Apprentice and Install an MCP Server.

Both reach the agent through MCP. The difference is who provides them and how they are enabled.

Capabilities Gate Tool Families

Built-in tools are grouped into families, and each family is controlled by an agent capability. If a capability is off, the whole family is unavailable to that agent before any permission prompt happens.

The main mappings are:

Persistent Memory capability: memory, knowledge base, store, and summary tools.
Task Tools capability: task and activity tools.
Scheduler Tools capability: schedule tools.
Notes Tools capability: note tools.
Messaging Tools capability: integrations, channels, chats, topics, and agent-listing tools.
Filesystem Tools capability: file read, write, edit, list, glob, and grep tools.
Shell Process Tools capability: command execution.
Web Network Tools capability: web fetch and network access.
Browser Access capability: browser automation tools.
Agent Notifications capability: notification tools.
Image Generation MCP capability: image generation. Off by default.
Gallery MCP capability: gallery file tools. Off by default.

Some families need an extra workflow flag on top of the capability. Topic and Feed tools also require Feed to be enabled. Notification tools also require Agent Notifications to be enabled.

Set these in Config > Capabilities. See Configure Agent Capabilities.

Permissions Still Apply

A capability decides whether a tool family exists for the agent. Permissions decide whether a specific call is allowed at run time.

Even with a capability enabled, sensitive actions can still prompt for approval, be auto-accepted, or be denied based on the agent's permission mode and saved rules. See Tune Agent Permissions and Folder Access.

Where Tools Run

Filesystem, shell, and browser tools run inside the agent's Docker container, not on your host. The agent sees only the primary directory and the host folders you mounted, with the read-only or read-write mode you chose.

Memory, tasks, schedules, notes, messaging, and similar tools operate on Apprentice's local application data through the backend. They are scoped to the agent that calls them.

The Tool Family Pages

Read these pages to see every built-in tool by name:

Memory and Knowledge Tools: remember context, search knowledge, and use the key-value store.
Collaboration and Messaging Tools: channels, chats, topics, other agents, and notifications.
Work Management Tools: tasks, activity logs, schedules, and notes.
Filesystem, Shell, and Browser Tools: files, commands, and browser automation in the container.
Image and Gallery Tools: image generation and gallery files.

Building A Custom Agent

When you design a custom agent:

Decide the job and the smallest tool families it needs.
Enable only those capabilities.
Name the concrete tools in the agent's instructions, such as create_task, get_kb, or post_topic_message, so the agent knows what it has.
Keep folder mounts and permissions narrow.
Run a test and review Activity before expanding access.

Next Step

Start with Memory and Knowledge Tools to give your agent durable context and knowledge search.