Collapse navigation tree

Collapse

AI Agent Quickstart

This quickstart helps you build your first AI agent in Redpanda Cloud. You’ll create an agent that understands natural language requests and uses MCP tools to generate and publish event data to Redpanda topics.

The Agentic Data Plane is supported on BYOC clusters running with AWS and Redpanda version 25.3 and later.

After completing this quickstart, you will be able to:

  • Create an AI agent in Redpanda Cloud that uses MCP tools

  • Configure the agent with a system prompt and model selection

  • Test the agent by generating and publishing events through natural language

Prerequisites

  • A BYOC cluster (agents are not available on Dedicated or Serverless clusters)

  • AI Gateway configured with at least one LLM provider enabled (OpenAI, Anthropic, or Google AI)

  • Completed the Remote MCP Quickstart to create an MCP server with the following tools deployed:

    • generate_input: Generates fake user event data

    • redpanda_output: Publishes data to Redpanda topics

What you’ll build

An Event Data Manager agent that:

  • Generates fake user event data (logins, purchases, page views)

  • Publishes events to Redpanda topics

  • Understands natural language requests like "Generate 5 login events and publish them"

The agent orchestrates the generate_input and redpanda_output tools you created in the Remote MCP quickstart.

Create the agent

  1. Log in to the Redpanda Cloud Console.

  2. Navigate to your cluster and click Agentic AI > AI Agents in the left navigation.

  3. Click Create Agent.

  4. Configure basic settings:

    • Display Name: event-data-manager

    • Description: Generates and publishes fake user event data to Redpanda topics

    • Resource Tier: Select XSmall (sufficient for this quickstart)

  5. Select your AI Gateway and model:

    • AI Gateway: Select the gateway you configured (contains provider and API key configuration)

    • Provider: Select a provider available in your gateway (OpenAI, Anthropic, or Google)

    • Model: Choose any balanced model from the dropdown

  6. Write the system prompt:

    You are an Event Data Manager agent for Redpanda Cloud.

Your responsibilities:
- Generate realistic fake user event data
- Publish events to Redpanda topics
- Help users test streaming data pipelines

Available tools:
- generate_input: Creates fake user events (login, logout, purchase, view)
- redpanda_output: Publishes data to the events topic

When a user asks you to generate events:
1. Use generate_input to create the event data
2. Use redpanda_output to publish the events to Redpanda
3. Confirm how many events were published

Always publish events after generating them unless the user explicitly says not to.

Response format:
- State what you're doing before calling each tool
- Show the generated event data
- Confirm successful publication with a count

  7. Select MCP tools:

    • Click Add MCP Server

    • Select the event-data-generator server (created in the MCP quickstart)

    • Check both tools:

      • generate_input

      • redpanda_output

  8. Set execution parameters:

    • Max Iterations: 30 (allows multiple tool calls per request)

  9. Review your configuration and click Create Agent.

    A service account is automatically created to authenticate your agent with cluster resources. For details about default permissions and how to manage service accounts, see Service account authorization.

  10. Wait for the agent status to change from Starting to Running.

Test your agent

Now test your agent with natural language requests.

  1. In the agent details view, open the Inspector tab.

  2. Try these example requests:

    Generate and publish 3 events 
    Generate 3 user events and publish them to the events topic.

    The agent should respond with these steps:

    1. Call generate_input to create 3 fake user events.

    2. Call redpanda_output to publish them to the events topic.

    3. Confirm the events were published.

    You should see the agent’s reasoning and the tool execution results.

    Generate specific event types 
    Create 5 login events for testing and publish them to Redpanda.

    The agent understands the request requires login events specifically and generates appropriate test data.

    Generate events without publishing 
    Show me what 3 sample purchase events would look like, but don't publish them yet.

    The agent calls only generate_input and displays the data without publishing.

  3. Navigate to Topics in the left navigation to verify events were published to the events topic.

Iterate on your agent

Try modifying the agent to change its behavior:

  1. Click Edit configuration in the agent details view.

  2. Update the system prompt to change how the agent responds. For example:

    • Add constraints: "Never publish more than 10 events at once"

    • Change output format: "Always format events as a table"

    • Add validation: "Before publishing, show the user the generated data and ask for confirmation"

  3. Click Save to update the agent.

  4. Test your changes in the Inspector tab.

Troubleshoot

For comprehensive troubleshooting guidance, see Troubleshoot AI Agents.

Common quickstart issue:

Events not appearing in topic: Verify the events topic exists and review the MCP server logs for publishing errors.

Next steps

You’ve created an agent that orchestrates MCP tools through natural language. Explore more: