Waiting for engine...
Skip to main content

Using Boomi Companion to build an Agentstudio agent with Claude

This page walks you through building a Boomi Agentstudio agent using Boomi Companion, from your initial prompt to a deployed agent you can hold a conversation with. It uses the bc-agentstudio plugin, which packages the boomi-agentstudio skill for Claude.

If you have already used Boomi Companion to build an integration, the flow feels familiar. You describe what you want in plain language, Boomi Companion builds it against the platform, and you review the result in the Boomi console. Building an agent adds a few steps that are specific to Agentstudio, such as packaging and deploying the agent before you can run it.

Install the bc-agentstudio plugin​

Prerequisites​

  • Access to the Boomi Enterprise Platform with Boomi Agentstudio (Agent Garden) enabled for your account.
  • An Agent Garden environment with an attached runtime to use as the deployment target.
  • (Optional) Any external targets the agent's tools will reach. Configure these ahead of time:
    • A deployed Boomi integration process for an Integration tool.
    • A Boomi DataHub repository for a DataHub Query tool.
    • A REST endpoint for an API tool.
    • An MCP server for MCP tools.

Core workflow​

Step 1: Configure your workspace and credentials​

Boomi Companion reads your platform credentials from a local .env file in your workspace. The boomi-agentstudio skill reuses the same BOOMI_USERNAME, BOOMI_API_TOKEN, and BOOMI_ACCOUNT_ID values that boomi-integration uses, and it adds three Agent Garden host settings:

  • AGENTSTUDIO_DESIGN_URL - the Agent Garden Design host for your region. The Design API covers authoring: agents, tools, packages, and deployments.
  • AGENTSTUDIO_ENVIRONMENT_ID - the Agent Garden environment where you deploy and test agents.
  • AGENTSTUDIO_EXECUTE_URL - the Agent Garden Execute host for your region. Use the Execute API only to invoke a deployed agent, so this value is needed only when you are ready to run the agent.

If you do not have a .env file yet, ask Boomi Companion to help you set one up. Alternatively, run /boomi-env-setup-guide to walk through it interactively.

Step 2: Describe the agent​

Open Claude with the bc-agentstudio plugin installed and describe the agent in plain language. The more clearly you state the agent's purpose and how it should behave, the better the result. A useful prompt covers:

  • The objective - a single input with a clear statement of what the agent is for.
  • How it interacts - a conversational agent for free-form chat, or a structured agent that takes a defined input object and returns a defined output object.
  • What it does - the actions or lookups it needs, which Boomi Companion maps to tools.
  • How it behaves - its tone, and any topics, terms, or patterns it must refuse or block, which Boomi Companion maps to guardrails.

Describe the outcome you want rather than the internal structure; you do not need to specify how many tasks to create or which fields belong in the agent definition. Boomi Companion has the Agentstudio knowledge to translate your intent into a valid agent.

For anything beyond the simplest agent, enter Plan mode (Shift+Tab twice) before prompting so that Boomi Companion thinks through its approach before building. In Plan mode, Boomi Companion describes what it intends to create - the agent's objective and tasks, the tools it will author or reuse, the guardrails it will apply, and the environment it will deploy to - before it makes any changes. Review the plan to confirm it matches your intent, then let it proceed.

Step 4: Build and deploy the agent​

Boomi Companion creates the agent through the Agent Garden Design API, in the order the platform requires:

  1. Authors and installs any tools the agent needs. Agentstudio tools come in several types:

    • A prompt the agent runs.
    • An API tool that calls a REST endpoint.
    • An Integration tool that triggers a deployed Boomi process.
    • A DataHub Query tool that reads golden records.

    A newly authored tool starts as a draft. Because packaging requires every referenced tool to be already installed, Boomi Companion installs each tool it creates. If the agent should call an existing Boomi integration, Boomi Companion first builds and deploys that process using the boomi-integration skill, and the Integration tool simply references it.

  2. Creates the agent definition with its objective, tasks, and guardrails, attaching the installed tools to the tasks that use them. Boomi Companion adds any guardrails you request on top of Agentstudio's always-on protections. Every agent already screens for profanity, prompt-injection attempts, and harmful content, so your guardrails only need to cover your own policies, such as blocking specific topics or patterns.

  3. Packages the agent. A package is an immutable, versioned snapshot of the current definition. Packaging requires the agent to have at least one task and every referenced tool already installed; Boomi Companion confirms both before it packages.

  4. Deploys the package to an environment, which produces the deployment identifier you use to run the agent.

If the platform returns a validation error during any step, Boomi Companion reads the error, consults its Agentstudio reference material, applies a fix, and retries.

Deployment environment

Over the Agent Garden API, you can deploy and run an agent only in a Production-classification environment; the platform rejects any attempt to deploy to a Test environment.

Production classification is an Agent Garden environment tier, not an instruction to point your agent at live business systems. While you develop and test an agent, keep the tools and data it touches non-sensitive, in line with the general guidance to keep production assets and sensitive data out of your work with Boomi Companion.

Step 5: Review the deployed agent​

Open Boomi Agentstudio in the Boomi console to view the completed agent. The agent's objective, tasks, tools, and guardrails match what you described, and the agent shows an active deployment. An agent authored via the API remains subject to the same field limits enforced by the console. You can view and edit everything Boomi Companion built directly in the console, just as you can with an agent you created by hand there.

Step 6: Run a test conversation​

Ask Boomi Companion to run a test conversation to confirm the agent behaves as intended: Meaning that it uses the right tool for a request, stays within its guardrails, and returns the expected result.

Step 7: Iterate on the agent (optional)​

You can refine an agent with follow-up prompts, such as adjusting its objective, adding a tool, or tightening a guardrail.

On this Page