# AIMock — Mock LLM, MCP, A2A, Vector APIs

> AIMock is a mock server for AI app testing: it simulates LLM providers, MCP tools, A2A agents, and vector/search APIs so tests avoid real paid endpoints.

## Install

Save as a script file and run:

## Quick Use

1. Install / set up:
   ```bash
   npm install @copilotkit/aimock
   ```
2. Start / smoke test:
   ```bash
   npx -p @copilotkit/aimock llmock -p 4010 -f ./fixtures
   ```
3. Verify:
   - Set `OPENAI_BASE_URL=http://127.0.0.1:4010/v1` and confirm your app/test hits the mock server (no real API calls).

## Intro

AIMock is a mock server for AI app testing: it simulates LLM providers, MCP tools, A2A agents, and vector/search APIs so tests avoid real paid endpoints.

- **Best for:** teams writing agent tests who need deterministic fixtures, record/replay, and one-port mocks
- **Works with:** Node.js projects, test runners (Jest/Vitest), OpenAI-compatible SDKs via `OPENAI_BASE_URL`
- **Setup time:** 10 minutes

## Practical Notes

- Mocks many surfaces from one port (LLM APIs, MCP, A2A, vector/search/rerank, etc.)
- Supports 12 LLM providers (per project docs) and streaming behaviors for realistic tests
- GitHub stars/forks (verified): see Source & Thanks

AIMock pays for itself the first time you avoid a flaky test or a surprise token bill.

Use it in three steps:

1. Start `llmock` (fixtures-first) or the full suite via `--config`.
2. Point your SDK to the mock by setting `OPENAI_BASE_URL` (and a dummy key if the SDK requires it).
3. Record real traces once (optional), then replay deterministically in CI forever.

Two good guardrails:

- Keep fixtures in version control and review them like code.
- Add a CI check that fails if your tests accidentally talk to real endpoints (base URL not patched).

### FAQ

**Q: What does it mock?**
A: LLM APIs plus agent protocols like MCP/A2A and common infra like vector DBs, search, and rerank.

**Q: Can it record real responses?**
A: Yes — it supports record & replay so you can capture and replay fixtures deterministically.

**Q: How do I adopt it safely?**
A: Patch base URLs via env vars in tests and add a CI guard to prevent real endpoint calls.

## Source & Thanks

> Source: https://github.com/CopilotKit/aimock
> License: MIT
> GitHub stars: 581 · forks: 37

---

<!-- ZH -->

## 快速使用

1. 安装 / 设置：
   ```bash
   npm install @copilotkit/aimock
   ```
2. 启动 / 冒烟测试：
   ```bash
   npx -p @copilotkit/aimock llmock -p 4010 -f ./fixtures
   ```
3. 验证：
   - 设置 `OPENAI_BASE_URL=http://127.0.0.1:4010/v1`，确认应用/测试请求命中 mock（没有真实 API 调用）。

## 简介

AIMock 是一个用于 AI 应用测试的“一站式 mock 服务器”：可模拟 LLM provider、MCP 工具、A2A agent、向量/搜索 API 等，让测试不再直连真实付费端点并降低波动与费用。

- **适合谁：** 需要可复现 fixture、record/replay、并希望一端口统一 mock 的 Agent 测试团队
- **可搭配：** Node.js 项目、Jest/Vitest 等测试框架、通过 `OPENAI_BASE_URL` 对接 OpenAI 兼容 SDK
- **准备时间：** 10 分钟

## 实战建议

- 一个端口即可 mock 多类能力（LLM API、MCP、A2A、向量/搜索/重排等）
- 官方文档列出 12 个 LLM provider 支持，并可模拟 streaming 行为
- GitHub stars / forks（已核验）：见「来源与感谢」

AIMock 的收益非常直接：一次避免 flaky 测试或“意外烧 token”就值回票价。

建议按三步使用：

1. 先用 `llmock`（fixture 驱动）或用 `--config` 跑全套 mock。
2. 通过 `OPENAI_BASE_URL` 把 SDK 指向 mock（某些 SDK 仍需要一个占位 key）。
3. 需要时先录一次真实 trace，之后在 CI 里长期稳定回放。

两个实践护栏：

- fixture 进版本库，像代码一样审查。
- CI 加断言：如果测试误连真实端点（base URL 没 patch）就直接失败。

### FAQ

**它能 mock 哪些东西？**
答：LLM API + MCP/A2A 等协议，以及向量库/搜索/重排等常见基础设施。

**能录制真实响应吗？**
答：可以，支持 record & replay，把真实响应保存为可复现 fixture。

**怎么安全接入？**
答：测试里用环境变量 patch base URL，并在 CI 加护栏防止误连真实端点。

## 来源与感谢

> Source: https://github.com/CopilotKit/aimock
> License: MIT
> GitHub stars: 581 · forks: 37


---
Source: https://tokrepo.com/en/workflows/aimock-mock-llm-mcp-a2a-vector-apis
Author: Script Depot