Tools are the bridge between a model and your live TouchDesigner project. Tool-capable operators expose a GetTool() surface — the Agent gathers those definitions, sends them to the model, and routes tool calls back to the operator that owns them.

Wiring a Tool

1. 1 Place an Agent controllers/agent

   Configure its model from ChatTD defaults or its own Model page.

2. 2 Place a tool operator e.g. Search, Tool DAT

   Set it up so it works on its own first.

3. 3 Enable tools Agent Tools page

   Turn on Use Tools.

4. 4 Add the tool to the sequence Tools sequence

   Drag the tool operator onto a Tool OP field in the Agent's Tools sequence.

5. 5 Turn the slot on Active

   Set Active to On for normal access, or Off to keep the slot but skip it.

⤢ expand

1

2

3

1. 1
   Use Tools

   Master toggle. Tools are skipped while this is off.
2. 2
   Add a block

   Use the sequence controls to add more tool slots.
3. 3
   Drop operator

   Drag a GetTool-enabled operator into the Tool OP field.

How It Works

Agent Tool Call ⤢ expand

prompttool callresultfollow-up

● data

Conversation

◆ agent

Agent

■ tool

Tool LOP

● data

Response

When a model requests a tool, the Agent runs the tool handler and records the turn in its output tables. If Tool Follow-up Response is on, the Agent makes another model call with the tool result and returns a final response.

Tool Operators

LOPs ships with many tool-capable operators:

• Tool DAT — stage edits to Table and Text DATs with diffs, confirmation, and undo.
• Tool Parameter — expose selected TouchDesigner parameters for agent control.
• Search and Search Text — web and local keyword retrieval.
• Search RAG and Graph — semantic and graph-based retrieval.
• MCP Client — import tools from external MCP servers.
• MCP Server and Tool Manager — expose TD tools to external agents.
• Web Viewer — give an agent browser navigation and page inspection.
• Tool Monitor — let an agent inspect recent TouchDesigner activity and, with a vision-capable model, request a network screenshot.

Tool Budget and Modes

A Tool Turn Budget of 1 works for simple “call one tool, answer” flows. Raise it when the Agent needs to search, fetch, inspect, and then act across multiple steps.

Each Tools sequence slot has an Active menu. On exposes that operator’s tools to the Agent; Off leaves the slot in place but skips it.

Use Approval Mode when a workflow should pause before tool execution. Approval can apply to all tools, destructive tools, or unknown tools depending on the mode.

Parallel tool calls speed up independent requests when the provider supports them. Keep parallel off when tool order matters or when tools change project state.

CAUTION Scope mutating tools carefully

Tools that edit DATs, parameters, files, or the network should be scoped narrowly. Use read-only presets, confirmation dialogs, path restrictions, and parameter filters before giving broad write access.

Other Agent Operators

Voice Agent and Claude Code use the same Tools sequence pattern as Agent. Any tool wiring that works with Agent works with them too.

Outside agents can also call your tools through MCP Server and Tool Manager . See External Agent Control.

Debugging

If a model is not calling a tool:

1. Check the Agent’s Tools page — is Use Tools on?
2. Check the tool block state — is Active set to On?
3. Check the tool operator itself — many tools expose nothing until their index, key, server, or target is configured.

Use Tool Debugger or Tool Registry to inspect schemas, duplicate names, and assignment state.