How to Add an Integration Test for an API Route with Claude Code

Unit tests prove a function works in isolation; integration tests prove a route actually responds correctly when wired together. Claude Code can write a test that sends a request to your handler and checks the status, the response body, and any side effect like a database write. The key is to test through the real entry point, not a mocked stand-in.

What you need

• Claude Code in your backend project
• A test runner and an HTTP test helper (supertest, the framework's test client, or fetch against a test server)
• One route you want covered end to end

Step 1: Point Claude at the route and its dependencies

Name the route file and tell Claude how the app is started in tests. If there is a test setup file or an existing integration test, point at it so the new test follows the same bootstrapping.

Step 2: Review the request and assertions

Read the generated test. Confirm the payload is realistic, the status assertion is correct, and the side-effect check actually queries the store rather than trusting the response alone.

import request from "supertest";
import { app, db } from "./setup";

it("creates a user", async () => {
  const res = await request(app)
    .post("/api/users")
    .send({ email: "a@b.com", name: "Ada" });

  expect(res.status).toBe(201);
  expect(res.body.id).toBeDefined();

  const row = db.get("select * from users where id = ?", res.body.id);
  expect(row.email).toBe("a@b.com");
});

Step 3: Add the failure cases

A route test that only covers the happy path is weak. Ask Claude to add a rejection case, like a duplicate email or a missing field, and assert the correct error status.

it("rejects a missing email with 400", async () => {
  const res = await request(app)
    .post("/api/users")
    .send({ name: "NoEmail" });

  expect(res.status).toBe(400);
});

Step 4: Run and watch for shared state

Run the integration suite. The most common failure here is leftover state between tests, so confirm each test resets the database. If tests pass alone but fail together, that is the cause.

Result: a test that posts to the real route, checks the 201 and the persisted row, rejects a bad payload with 400, and runs cleanly thanks to a per-test database reset.