Test Your Agent

For the complete MongoDB documentation index, see www.mongodb.com/docs/llms.txt

Test Your Agent

MongoDB Atlas Agent Engine is in Public Preview and intended for evaluation and prototyping purposes only. We do not recommend production deployments during preview and expect to communicate breaking changes with you periodically. To learn more about preview feature considerations, see Preview Features.

Overview

In this guide, you can learn how to manually test your Atlas Agent Engine agent in a local development environment. After you start the environment by using the agentengine dev command, the CLI prints URLs for the app and the local MongoDB instance so that you can drive requests, inspect responses, and debug issues before you build and deploy the agent to the cloud.

Prerequisites

Before you begin, ensure that you complete the following tasks:

  • Install and configure the agentengine CLI.

  • Start a local development environment by following the steps in the [Run and Test Your Agent Locally](https://www.mongodb.com/docs/agentengine/build/run-local.md#std-label-agent-engine-build-local guide.

To learn more about how to install the CLI, authenticate, and register the project, see [Install and Authenticate.](https://www.mongodb.com/docs/agentengine/build/install-authenticate.md#std-label-agent-engine-set-up-install-authenticate

Access Local Services

When you run agentengine dev up in either hot-reload or isolated mode, the CLI builds the Docker image for your agent and starts the full stack. When all services are ready, the CLI prints URLs for the app and the local MongoDB instance.

Use the following URLs to perform your manual tests:

  • App endpoint URL: Use a browser, API client, or command-line tools to send requests to your agent.

  • MongoDB Atlas Local URL: Connect by using mongosh or another MongoDB client to inspect or seed test data.

For example, you can use these URLs to perform the following tests:

  • Send a chat or HTTP request that exercises a specific tool call.

  • Insert test documents into the local database and verify how the agent uses them.

  • Inspect responses for correctness, latency, and error handling.

Test in Hot-Reload Mode

Hot-reload mode mounts your source code into the app container and reloads your agent process each time you save a change. Use the following workflow to test and iterate without rebuilding the stack.

To learn how to start the environment in hot-reload mode or attach VS Code to the running container, see [Hot-Reload Mode (Recommended)](https://www.mongodb.com/docs/agentengine/build/run-local.md#std-label-agent-engine-build-local-hot-reload in the Build Your Local Environment guide.

  1. Exercise the agent.

Use the app or playground UI URL to send requests that cover the behavior you want to test.

  1. Edit the agent code.

Change your Python code, prompts, or tool wiring in your project source. The file watcher inside the container detects changes and reloads the affected services without a full rebuild.

  1. Re-run the same scenario.

Re-send the same request and verify that your changes produce the expected behavior.

Test in Isolated Mode

Use isolated mode when you want to validate behavior that depends on service boundaries, container restarts, or network interactions.

To learn how to start the environment in isolated mode, see [Isolated Mode](https://www.mongodb.com/docs/agentengine/build/run-local.md#std-label-agent-engine-build-local-isolated in the Build Your Local Environment guide.

  1. Run end-to-end scenarios.

Use the app and playground UI URLs to exercise your agent as you would in production.

  1. Apply code changes.

Because isolated mode does not mount your source code from the host, you must rebuild images after code changes. Run the following commands to apply your changes:

sh agentengine dev stop agentengine dev up --isolated

The first command stops the running containers. The second rebuilds and restarts the isolated stack with your updated code.

Inspect Logs and Troubleshoot

While you test, use agentengine dev commands to inspect logs and reset state as needed.

View Logs

To stream logs from all running services, run the following command:

agentengine dev logs

Use this output to debug errors, verify tool calls, and inspect trace messages emitted by your agent.

Reset Local State

If you encounter persistent errors related to the local MongoDB replica set or data, you can reset the local state by running the following commands:

agentengine dev clean
agentengine dev up

This sequence removes all local volumes and starts a fresh stack.

Warning:

Running agentengine dev clean permanently deletes all local data stored in the MongoDB Atlas Local volume. Back up any data you need before you run this command.

Next Steps

After you validate your agent locally, you can build and deploy it to the Atlas Agent Engine cloud environment. To learn how to package your agent image and deploy it to a workspace, see the [Deploy the Agent](https://www.mongodb.com/docs/agentengine/deploy/provision-secrets.md#std-label-agent-engine-deploy-provision-secrets section.