> Call the `diagnose` tool first. Full index: /llms.txt. Errors: /reference/problems.

# Projects and apps

A **project** holds everything for one product: its apps, scenarios, test plans, test data, cards and runs. Every MCP call names its project with `project_id`.

An **app** is what a run tests. TaloTrace supports two kinds:

| App         | What you give TaloTrace                                                                    | Where it runs                     |
| ----------- | ------------------------------------------------------------------------------------------ | --------------------------------- |
| Web app     | A start URL: a public HTTPS page, or a page on your laptop such as `http://localhost:3000` | A real browser in the cloud       |
| Android app | An uploaded build (an APK), or a live phone that runs your working tree                    | A real Android phone in the cloud |

iOS is not available.

## Create them

In the web app, create a project and add an app. Your agent can do the same with the `setup` tool: `list_projects`, `create_project`, `list_apps` and `create_app` (use `list` first so nothing is duplicated). See [Setup and diagnose](/reference/mcp/setup-diagnose) for the arguments.

## Start URL

A web app has a start URL, the page where every run begins. For a page on `localhost`, your agent passes the start URL itself and the CLI opens a tunnel for the run. See [Test an app on localhost](/web/test-localhost).

## Test accounts

If the app has a sign-in, add a test account for it (`setup create_account`, or **Test data** in the web app). Use an account made for testing. TaloTrace and your agent never invent credentials.