Attach tools
By the end of this page you will know how to give a skill a set of tools, configure their defaults, and test them in place.
Prerequisites
- A skill added to your agent (Skills overview).
- Admin or Developer role.
- Familiarity with Tools — system, project, and MCP tools.
How a skill’s tools work
Each skill has an optional tool whitelist. When the skill is active, only its whitelisted tools are available to the agent.
Leave the tool list empty to let the agent use all of its available tools when the skill is active. Set a whitelist only when you want to scope the skill to specific tools.
Select tools
Open the tool picker
In the skill card’s Tools section, click Add Tools (or Add More if some are already selected). The Select Tools dialog opens, subtitled “Choose which tools this skill can use when active.”
Pick from the four groups
Tools are grouped into four sections, each with its own selection count and a shared search box:
| Group | What it contains |
|---|---|
| System Tools | Built-in tools (calculator, web search, database, RAG, and more). See Built-in tools. |
| Project Tools | Custom HTTP and Python tools you built. See Tools. |
| MCP Tools | Tools discovered from connected MCP servers, grouped by server. See MCP servers. |
| Task Agents | Your Task Agents, usable as tools by the AI Agent. |
MCP tools that run on a remote MCP server show a Remote badge. A Task Agent shows a Form badge when it has an initiation form.
Confirm your selection
Search or scroll to find tools, tick the ones you want, and click Confirm Selection. The dialog footer shows the running count of selected tools.
Attaching a Task Agent as a tool lets your AI Agent call a deterministic, multi-step workflow when it needs one — for example, a “process refund” pipeline — while the AI Agent handles the open-ended conversation around it.
Review selected tools
Back on the skill card, selected tools appear as badges. Task Agent tools show a bot icon. A configured tool shows a filled dot. Each tool badge has:
- A gear button — opens Configure & Test for that tool.
- An × — removes the tool from the skill.
The section header shows the count and offers Add More and Clear All.
Configure and test a tool
Click the gear on a tool to open the Configure & Test dialog. It has two tabs.
Config tab
Set default parameters for the tool — values that are pre-filled on every call. Values the LLM provides at run time take precedence over your defaults.
You can also override Human-in-the-Loop (HIL) for this tool within this skill:
- Default — use the tool’s own HIL setting.
- On — always require approval before this tool runs.
- Off — run the tool directly, without an approval prompt.
Test tab
Run the tool with sample inputs and inspect the result, so you can confirm the configuration works before relying on it.
Save your changes
Click Save in the header to persist the skill’s tools and per-tool configuration into a draft version.
Common issues
- The agent can use tools you didn’t expect — The skill’s tool list is empty, so all agent tools are allowed. Add a whitelist to scope it.
- A tool needs credentials or setup — Some system and project tools require configuration or secrets before they work. See Configuration & secrets.
- An MCP tool is missing from the picker — Its server may be disconnected or its tools disabled. See Manage an MCP server.