Switching from MSW to aimock

MSW is excellent for general API mocking. But when your AI app spans multiple processes—Playwright, Next.js, agent workers, Docker—MSW's in-process interception can't reach them. aimock runs a real HTTP server that any process can hit.

The 5-minute switch

Side-by-side: streaming SSE with MSW vs. aimock.

MSW — streaming SSE (~35 lines) ts

import { http, HttpResponse } from 'msw'
import { setupServer } from 'msw/node'

const server = setupServer(
  http.post('https://api.openai.com/v1/chat/completions', () => {
    const encoder = new TextEncoder()
    const stream = new ReadableStream({
      start(controller) {
        controller.enqueue(encoder.encode(
          'data: {"choices":[{"delta":{"role":"assistant"}}]}\n\n'
        ))
        controller.enqueue(encoder.encode(
          'data: {"choices":[{"delta":{"content":"Hello"}}]}\n\n'
        ))
        controller.enqueue(encoder.encode(
          'data: {"choices":[{"delta":{"content":" there"}}]}\n\n'
        ))
        controller.enqueue(encoder.encode('data: [DONE]\n\n'))
        controller.close()
      }
    })
    return new HttpResponse(stream, {
      headers: { 'Content-Type': 'text/event-stream' }
    })
  })
)
server.listen()

aimock — same result (4 lines) ts

import { LLMock } from '@copilotkit/aimock'

const mock = new LLMock()
mock.onMessage("hello", { content: "Hello there" })
await mock.start()
// set OPENAI_BASE_URL = mock.url + "/v1"

Non-streaming comparison

For simple JSON responses, MSW is comparable:

MSW ts

http.post('/v1/chat/completions', () => {
  return HttpResponse.json({
    choices: [{ message: { content: "Hi" } }]
  })
})

aimock ts

mock.onMessage("hello", { content: "Hi" })

For non-streaming responses, the complexity is similar. aimock's advantage is consistency—the same fixture works for streaming, non-streaming, and WebSocket endpoints.

What you gain

🌐

Cross-process interception

Playwright → Next.js → agent workers → Docker. Every process on the machine hits the same mock.

⚡

Built-in SSE for 10+ providers

OpenAI, Claude, Gemini, Bedrock, Azure, Vertex AI, Ollama, Cohere. No manual chunk construction.

🔌

WebSocket APIs

OpenAI Realtime, Responses WS, Gemini Live. MSW cannot intercept WebSocket.

⏺

Record & Replay

Proxy real APIs, save as fixtures, replay forever. npx aimock --record --provider-openai https://api.openai.com

📁

Fixture files

JSON files on disk. Version-controlled. Shared across tests.

🧩

MCP + A2A + Vector

Mock your entire AI stack, not just LLM calls.

What you keep (or lose)

Capability	MSW	aimock	Notes
Browser service worker	✓	✗	Point browser app's API base URL at aimock instead
General REST/GraphQL mocking	✓	✗	Keep MSW alongside for non-AI routes
Cross-process	✗	✓	aimock's key advantage
Streaming SSE	Manual	Built-in	10+ providers
WebSocket	✗	✓	3 protocols
Record & replay	✗	✓
Chaos testing	✗	✓
Zero dependencies	✗	✓	MSW ~300KB

Using aimock alongside MSW

You don't have to choose. Use MSW for general REST/GraphQL mocking and aimock for AI-specific endpoints.

test-setup.ts ts

// test setup
import { setupServer } from 'msw/node'
import { LLMock } from '@copilotkit/aimock'

// MSW for REST APIs
const mswServer = setupServer(
  http.get('/api/users', () => HttpResponse.json([...]))
)

// aimock for AI APIs
const aiMock = new LLMock({ port: 0 })
aiMock.onMessage("hello", { content: "Hi!" })

beforeAll(async () => {
  mswServer.listen()
  await aiMock.start()
  process.env.OPENAI_BASE_URL = aiMock.url + "/v1"
})

CLI / Docker quick start

CLI sh

npx aimock -p 4010 -f ./fixtures

Docker sh

docker run -d -p 4010:4010 \
  -v $(pwd)/fixtures:/fixtures \
  ghcr.io/copilotkit/aimock:latest \
  -p 4010 -f /fixtures