Tools

Tools are the actions an agent can take at run time: searching the web, reading a file, remembering a fact, calling another agent, and more. The Tools tab in the Library is the catalog of what is available in your active workspace.

Use tools when an agent needs to do something rather than only produce text. You attach tools to an agent in its definition (see Agents overview) or wire them into a graph with a Tool node.

Bundled native tools

These tools ship with the runtime and are available out of the box.

Tool	What it does
filesystem	Read and write files (within sandbox limits)
web	Web search and fetch — Tavily-backed search via MCP
memory	Store and recall facts across turns
time	Current time and date helpers
process	Run and manage subprocesses
knowledge	Query the workspace knowledge / context store
tasks	Create and track tasks
agent_find	Locate other agents available to call
consult	Invoke another agent (agent ↔ agent)

How tools reach agents

A tool in the catalog can come from three sources:

• Bundled natives — the built-in tools listed above.
• Skills — any skill that exposes a callable interface appears here as a tool.
• MCP servers — tools exposed by connected MCP servers, including Tavily web search.

Whatever the source, an agent calls all of them through the same tool interface.

The consult tool

The consult tool lets one agent call another agent and use its answer — the foundation of multi-agent panels and orchestration. Because agent-to-agent calls can fan out, consult is governed by communication policies: an agent can only consult the agents its policy allows.

Note

See Communication policies for how to control whether and how agents may consult one another.

Adding tools

1. To add a bundled tool, attach it to your agent in the agent definition or drop a Tool node into a graph and select it.

2. To add a skill-backed tool, create and publish a skill — it then appears in the Tools catalog.

3. To add an external tool, connect an MCP server; its tools join the catalog automatically.