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

# Test localhost with Claude Code

You need a web app that runs on your laptop, Claude Code, and Node 22.14 or newer. The app keeps its `localhost` address: TaloTrace reaches it through a tunnel that your agent opens for you.

#### With your agent

1. Start your app, for example on `http://localhost:3000`.

2. In your project folder, run the one-command setup. It signs you in with tool access and installs the skills and the MCP entry. See [One-command setup](/install/setup-command).

   ```bash
   npx -y @talotrace/connect setup
   ```

3. Restart Claude Code in the folder so it loads the new skills and the `talotrace` MCP server.

4. Paste this sentence to the agent:

   ```text
   Verify this change with TaloTrace
   ```

5. The agent calls `diagnose`, finds or creates your project and app, picks a [scenario](/concepts/scenarios-and-test-plans) that covers the change (it writes one only when none does) and starts a run on `http://localhost:3000`.

6. Wait for the report. It names the scenario, the run id, the verdict, a link to watch the run and, on a failure, the cause.

#### From a terminal

1. Sign in with tool access:

   ```bash
   npx -y @talotrace/connect login --tools
   ```

2. Check the machine and your app's port:

   ```bash
   npx -y @talotrace/connect doctor --ports 3000
   ```

3. Gate one scenario and wait for the verdict:

   ```bash
   npx -y @talotrace/connect check --project <project id> --app <app id> --scenario <scenario id> --start-url http://localhost:3000/ --ports 3000 --timeout 10m --evidence-dir ./evidence
   ```

   The exit code is the verdict. `0` means passed. See [exit codes](/reference/cli/exit-codes).

#### In the web app

Open [app.talotrace.ai](https://app.talotrace.ai), create a project and add your app. A localhost page cannot be reached from the web app alone, so keep `talotrace connect 3000` running in a terminal while a run uses it. See [Test an app on localhost](/web/test-localhost).

## Read the verdict

| Verdict      | What it means for you                                                             |
| ------------ | --------------------------------------------------------------------------------- |
| `passed`     | The run ended, every step executed and the analysis found no failure.             |
| `failed`     | The run ended and the analysis found a failure. Read the cause and apply the fix. |
| `incomplete` | There is no usable result. Treat the change as unverified, never as passed.       |

After a fix, ask the agent for one rerun and a comparison. See [Prove a fix](/guides/prove-a-fix).

## If it does not work

* Run `talotrace doctor`. See [Check your install](/check-your-install).
* A refused call carries a code. Look it up on the [problems page](/reference/problems).
* `INSUFFICIENT_SCOPE` means the sign-in has no tool access. Run `talotrace login --tools`.