Agent Studio

Learn how to use Agent Studio to build your agents.

You build and test agents using Agent Studio, Toolhouse's integrated agent development environment.

Agent Studio allows you to build, run, test, and deploy agents with ease.

You build agents in Agent Studio using only a prompt, and Toolhouse automatically connects them to the Toolhouse infrastructure. Based on your prompt, Toolhouse selects the right MCP servers to allow your agents to perform required actions such as retrieving knowledge (RAG), storing and retrieving user information, browse the web, and more. All the underlying prompts, code, and the overall infrastructure is managed for you by Toolhouse. You can test and edit your prompts, and run them as many times as you like.

Try an agent on Agent Studio

You can share Agent Studio agents, or keep them private. This is useful for sharing and editing agents when you work with your team. You can also clone existing agents so you can fork existing prompts while keeping track of all the different versions.

Once you're happy with the results of your agent, you can automatically deploy it as an API using Agent Runs, or schedule it to run at a particular cadence.

Access Agent Studio

  1. On the sidebar of the Toolhouse dashboard, find the Agent Studio option and click on it.

  2. Enter your prompt or use a Agent Studio template by clicking on the suggestions.

The Toolhouse MCP server is installed by default

Based on your prompt, Toolhouse selects the right MCP servers to allow your agents to perform required actions. You can also explicitly ask the agent to use specific servers by using Bundles.

You can make the agent use a Bundle you previously created. To do so, simply select one of your existing Bundles from the left-hand sidebar.

Running an Agent

To run an Agent, simply click Run.

You will be able to inspect the agent's progress as it runs.

  • Agent Steps will show any MCP servers called, as well as intermediate reasoning steps, if available.

  • Agent Output will stream as soon as the agent produces its final output. Use this output to determine whether the agent performed the tasks correctly.

You can run the agent again by clicking Run again. This has the same effect as clicking Run.

When you are happy with the output, click Continue.

Deploying the Agent

Clicking Continue will make the Agent available as an API. You will see a section called Do this with the result.

  • You can run the Agent as an API using Agent Runs.

  • You can also schedule the agent to run at particular cadences. Note that you can also do this via the API.

Share an Agent

You can share your Agent Studio agents with others. Shared agents are accessible to anyone with the link. This is useful when working with teams, or when you want to showcase interactions and behaviors of your agents.

You can share agents by simply copying and sharing the URL of the agent. You can also click the Share button on the page.

Only the owner can change the original agent

Only the agent owner has the ability to edit the agent behavior within the original URL. Other users will view the agent in read-only mode; they can clone the agent for use in their own Toolhouse account, where they can continue editing with their own prompts.

Running a shared Agent

To run an Agent shared with you, simply click Run this agent on top of Agent Studio. This action will fork the original agent and you will be able to change it and run it without affecting the agent you're cloning.

Bundles may affect how the agent behaves

Note that some agents may rely on MCP servers configured via Choose MCP servers for your agent. Bundles are personal to the author of the chat and may contain servers that require configuration, such as API keys. For this reason, Bundles are not shared when you clone a shared agent.

If a cloned agent is not behaving as intended, ensure you have a Bundle containing the right MCP servers with the correct configuration.

Change access settings

Agent Studio agents are public, meaning they are available to anyone with the link. To make an agent private, simply click on the Public button and select Private.

Last updated