Configuring Tools

Tools are one of the main primitives for building agentic workflows & systems. Agents can choose to call tools as they see fit in order to learn about their environment, and to take actions to act upon this information. Typically, tools will be accessed through an MCP server. To learn more about tools, see here.

Configuring tool parameters

Every tool exposes input parameters. When you add a tool to an agent, a configuration modal lists each parameter with a Scope dropdown. The scope decides where the value comes from at runtime. This is the single most important choice when configuring a tool, and it is the same for every tool in Agent Studio.

There are three scopes.

Scope	Who sets the value	When it is set	Use it for
Fixed	You, during configuration	Once, at setup	Values that never change between runs
User input	The person (or system) running the agent	Each time the agent is invoked	Values that change per run
Agent determined	The agent (LLM)	During the run, by reasoning	Values the agent should figure out itself

Fixed

You set the value once while configuring the agent, and it cannot change at runtime. Choose Fixed for settings that should be the same on every run.

• Example: pin marketplace_id to a specific marketplace so every search targets it.
• Example: set a limit of 20 so results are always capped the same way.

User input

The value is supplied when the agent runs. If the agent’s input schema already has a field with the same name and a compatible type, the parameter maps to it automatically. If not, Agent Studio adds the field to the agent’s input schema when you save. Choose User input for values that vary from run to run but should not be left to the agent to guess.

• Example: data_product_id, so the caller picks which data product to query.
• Example: a date range that the requester provides each time.

Agent determined

The agent decides the value at runtime based on the conversation and its reasoning. Choose Agent determined for parameters that require inference.

• Example: the sql parameter on a SQL tool, which the agent writes from the user’s question.
• Example: result_table_name, a descriptive label the agent generates.

Note

Some parameters are restricted from Agent determined scope for security or consistency. For those, you must use Fixed or User input.

Tip

Fixed and User input values are passed straight to the tool and do not appear in the agent’s context. Agent determined values are generated by the model during the run and show up in its reasoning trace, which you can inspect when testing an agent.

Each tool reference page lists its parameters and notes a sensible default scope for each one. See Tool input configuration for where this appears in the agent builder.