
Smolanalytics
io.github.Arjun0606由 GitHub 账号 Arjun0606 发布v0.92.1更新于 Oct 24, 2026
End-to-end tests from one sentence: an agent drives a real browser, and passing runs replay free.
概览
让助手用一句自然语言编写并运行端到端浏览器测试,通过的运行可回放且不再调用模型。
- 可以做什么
- 该服务器提供三个工具:run_tests、can_i_ship 和 list_tests。测试是 markdown 文件中的一句话;代理会打开真实浏览器,通过无障碍树读取页面,推断如何执行该句描述的检查,并返回五种判定之一:passed、failed、stale、errored 或 flaky。通过的运行会被录制,之后的运行回放该录制且不调用模型,直到录制不再匹配应用。
- 适用场景
- 适合希望为 Web 应用做端到端检查、又不想编写和维护基于选择器的测试代码的场景,也适合让助手运行一组句子测试并报告哪些失败。可用于本地或 CI 中针对预览地址运行,包括在每个拉取请求上留下评论。
- 运行要求
- 通过 npx 或 OCI 镜像作为本地进程运行,需要 Node.js 或容器运行时。首次运行尚无录制的测试需要你自己的 ANTHROPIC_API_KEY;回放不需要密钥。需要访问目标站点和模型提供方的网络。托管服务是另外的付费产品。
安装
在 SourceWeft 中
- 打开 控制台中的 Smolanalytics,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
smolanalytics
End-to-end tests written as one English sentence. An agent runs them in a real browser, and a passing run replays with no model call.
A test is one sentence in a markdown file. An agent opens a real browser, reads the page through its accessibility tree, works out how to do what the sentence says, and returns a verdict. There is no test code, so there is no selector to update when a button moves.
A run that passes is recorded, and every run after it replays that recording with no model call at all. The agent comes back only when the recording stops fitting your app.
Five verdicts, never blurred: passed, failed, stale, errored, flaky. Only failed is a claim about your app. A recording that stopped fitting is stale, never a bug report, and a runner that could not finish is errored, never counted against you.
On every pull request
test --suite tests/ --url <preview> --comment runs a folder of sentences and leaves one comment on
the pull request, edited in place on every push: what broke, the file most likely responsible and
the evidence for that guess. Exit codes are 0 (clean), 1 (your app failed a test) and 2 (the runner
could not finish). The workflow file is in the docs.
In your editor (MCP)
Three tools: run_tests, can_i_ship and list_tests. Listed on the official MCP Registry as
io.github.Arjun0606/smolanalytics.
Evidence rather than claims
smolanalytics.com/proof is a GitHub Action that runs this product's own suite against two live sites every night and publishes every verdict, including the red ones. Nothing on that page is written by hand.
What it costs
The CLI runs on your machine or your CI runner, on your own ANTHROPIC_API_KEY. A first run of a
test costs roughly what one agent conversation costs; a replay costs nothing. The hosted side at
smolanalytics.com is $19/month with 100 tested pull requests included,
after a 14-day trial with no card. This is commercial software, not open source.
Full CLI reference: cli/README.md. Questions: smolanalytics.com/contact.
来源:README.md,提交 38bc278
工具
0版本历史
1- v0.92.1最新Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分