Skip to main content
Tool Exposure controls how your agent’s or gateway’s tools are presented to the consuming AI. As you attach more capabilities, the number of available tools grows — Tool Exposure lets you manage that complexity.

​
Configuration

1

Open Settings

Agents > Select agent > Settings (or Gateways > Select gateway > Settings)
2

Find Tool Exposure

Locate the Tool Exposure section
3

Choose a Mode

Select Smart (recommended), Omni, or Individual
4

Save

Click Save

​
Visibility Modes

​
How Each Mode Works

Exposes 3 meta-tools for progressive discovery:
  1. discover_tools — List available tools matching a task description
  2. get_tool_schema — Get detailed schema for a specific tool
  3. execute_tool — Run a tool with specified parameters
The AI first discovers what’s available, inspects the right tool, then executes it. This gives the AI context to make good choices without overwhelming it with every tool upfront.

​
Omni Mode

Exposes a single omni_tool. The AI describes what it wants in natural language, and an LLM-powered discovery layer finds and executes the right tool automatically.

​
Individual Mode

Exposes every tool directly. The AI sees the full list and picks which to call. Best when you have a small, well-defined set of tools or need precise control.

​
Confidence Scoring (Omni Mode)

In Omni mode, the discovery LLM rates match confidence (0.0–1.0):
  • 0.95+ — Very confident (clear match)
  • 0.80–0.95 — High confidence (strong match)
  • 0.70–0.80 — Moderate (passes default threshold)
  • Below 0.70 — Rejected, asks for clarification
The confidence threshold is a global setting (default: 0.7). Below the threshold, Omni Tool asks the user to clarify instead of guessing wrong.

​
Which Mode to Choose

​
Applies to Both Agents and Gateways

Tool Exposure is configurable on both agents and MCP gateways. The behavior is the same — it controls how tools are presented to the consuming AI client.

​
Cost

For current pricing details, see Pricing. In Smart and Omni modes, tool discovery adds a small LLM call. You only pay for the actual tool execution.

​
Troubleshooting

​
Next Steps