Skip to content
See which Tools a Task can use
Public Preview

Task Tools: Built-in, SDK, and MCP

Explains how Tasks use Built-in, SDK, and MCP Tools, and how to choose between an SDK Tool Host and an MCP Connection.

This page covers which Tools a Task can use, Built-in Tools, choosing between SDK Tools and MCP Connections, and what the model reads from a Tool.

  • Tools available to a Task and its Agent Identity.
  • Built-in Tools: Library search, Web research, Code Execution, Recall, and Delegated work.
  • Choosing between an SDK Tool Host and an MCP Connection.
  • What the model reads from an offered Tool.

Summaries were generated by AI.

A Task gets Built-in Tools that Duale AI supplies, SDK Tools that your application serves, and MCP Tools from a remote server that you connect.

See which Tools a Task can use

A Task and each of its continuations use the current Tools of their Agent Identity. A change on a Tool Host or an MCP server can add, remove, or rename a Tool while the Task runs. A Task request can neither add nor narrow these Tools: to change what a Task can use, change what its Agent Identity can reach. To limit what one kind of work can reach, run it under an Agent Identity that has only the Tool Hosts and MCP Connections that work needs.

Built-in Tools

Duale AI supplies these Built-in Tools to every Task, and no setting removes them. You do not register, assign, or call them; the model requests them when the work needs them.

Built-in ToolWhat a Task can do with itWhat you controlRead more
Library searchFind, match, list, and read documents in the Libraries its Agent Identity can reachWhich Libraries the Agent Identity can reachWhat agents can do with Library tools
Web researchSearch the public web and read public pagesWhat the instruction asks forWeb research
Code ExecutionRun Python or JavaScript on the Task’s machine, and copy Library documents into itWhat the instruction asks for, and which Libraries the Agent Identity can reachCode Execution
RecallReread its own earlier work, and search earlier Tasks of the same Agent IdentityWhich Agent Identity runs the TaskContext continuity
Delegated workDelegate focused parts of its work to child Executors inside the same TaskThe instruction and the deadlineHow a Task divides its work
  • Built-in Tool
    Library search
    What a Task can do with it
    Find, match, list, and read documents in the Libraries its Agent Identity can reach
    What you control
    Which Libraries the Agent Identity can reach
  • Built-in Tool
    Web research
    What a Task can do with it
    Search the public web and read public pages
    What you control
    What the instruction asks for
    Read more
    Web research
  • Built-in Tool
    Code Execution
    What a Task can do with it
    Run Python or JavaScript on the Task’s machine, and copy Library documents into it
    What you control
    What the instruction asks for, and which Libraries the Agent Identity can reach
  • Built-in Tool
    Recall
    What a Task can do with it
    Reread its own earlier work, and search earlier Tasks of the same Agent Identity
    What you control
    Which Agent Identity runs the Task
  • Built-in Tool
    Delegated work
    What a Task can do with it
    Delegate focused parts of its work to child Executors inside the same Task
    What you control
    The instruction and the deadline

Choose between an SDK Tool and an MCP Connection

Your own Tools come from an SDK Tool Host or an MCP Connection.

OptionUse it whenWho runs the ToolHow its Tools stay available
SDK Tool HostYou need custom code, control over when Tool definitions change, or your own approval stepYour application processThe SDK publishes the Tool definitions for one Agent Identity. Run and change the Tool Host states how long they stay offered
MCP ConnectionAn existing remote MCP server already exposes the capabilities you needThe remote MCP serverThe server’s compatible Tools are offered to every assigned Agent Identity while catalog checks succeed
  • Option
    SDK Tool Host
    Use it when
    You need custom code, control over when Tool definitions change, or your own approval step
    Who runs the Tool
    Your application process
    How its Tools stay available
    The SDK publishes the Tool definitions for one Agent Identity. Run and change the Tool Host states how long they stay offered
  • Option
    MCP Connection
    Use it when
    An existing remote MCP server already exposes the capabilities you need
    Who runs the Tool
    The remote MCP server
    How its Tools stay available
    The server’s compatible Tools are offered to every assigned Agent Identity while catalog checks succeed

What the model reads from a Tool

SDK Tools and MCP Tools reach the model in the same form. For each Tool offered to it, the model reads:

  • the model-facing name of the Tool;
  • a name that tells where the Tool comes from, such as the MCP Connection name, then the Tool’s description;
  • the parameter schema, with each description and constraint inside it.

The model reads nothing else from a Tool, and uses these to choose a Tool and write its arguments. An offered Tool’s current definition can steer the model even when the model does not call that Tool.

Give every Tool a distinct name and a description, so that the model can choose the right Tool when an Agent Identity has many Tools. Put what the Tool does and when to use it first, because the model reads only the start of a long description.

Next steps