Ruin Everything

MCP开发工具

提供方 未标注

开发工具 HTTP · 远程 会话登录 社区 / 第三方 ClaudeCursorVS CodeGemini CLI
查看接入方式

检查本地 Chrome 中的应用标签页,并通过职位 URL 获取已保存的 Ashby 提交回执。

一句话选型结论

按资料字段整理

Ruin Everything 属于开发工具类 MCP 服务器,HTTP · 远程,会话登录,可直接接入 Claude、Cursor、VS Code 等,有可直接复制的远程端点配置。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类开发工具 传输方式HTTP · 远程 鉴权会话登录 部署方式支持远程部署
GitHub 星标 未标注 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

开发工具
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

开发

安装与配置

HTTP · 远程 会话登录
连远程 HTTPS 端点,通常为 Streamable HTTP
uvx --from 'ruin-everything @ git+https://github.com/yuki4266/ruin-everything.git@main' ruin browser

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    确认你的客户端支持远程端点,并拿到服务地址。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
代码与仓库操作

在对话里直接查代码、提 issue、走 PR,省掉来回切工具。

本地命令与脚本

把构建、测试、格式化这类重复命令交给模型执行。

接口联调与测试

让模型调你的接口、跑一遍用例,快速定位问题。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude、Cursor、VS Code,换客户端不用重新研究配置。

先想清楚的地方

  • 远程端点依赖网络与服务方可用性,注意密钥与调用额度。
  • 使用前要准备好会话登录对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在开发工具场景里让模型直接调用外部能力的人。
  • 已经在用 Claude、Cursor 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Ruin Everything当前 未标注 HTTP · 远程 支持远程部署 有 2026-11-20
Git 838e2ec1 12 stdio · 本地进程 本地部署 有 2026-05-17
Markitdown MCP 2c653a5f 1 stdio · 本地进程 本地部署 有 2026-05-17
JavaGuide 未标注 stdio · 本地进程 本地部署 有 2026-04-11
awesome-claude-skills 未标注 stdio · 本地进程 本地部署 有 2026-04-26
Context7 831acf40 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

开发工具
字段来自项目仓库与公开索引,中文说明由本站标注
提供方未标注上游标注的维护方 未标注
分类开发工具本站按上游分类展示
连接方式HTTP · 远程连远程 HTTPS 端点,通常为 Streamable HTTP
鉴权方式会话登录远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型支持远程部署上游给出的部署形态
最近核验2026-11-20上游最后核验时间

详细介绍

12.8 KB
来自项目仓库与上游目录的原文,未做删改

English · 简体中文 · 日本語 · Español

Ruin Everything

Give your agent the browser. Get your time back.

Your agent clicks. Ruin keeps the work moving.

Ruin Everything is a local Chrome runtime for browser agents. It lets your agent use the sessions, cookies and extensions already in your shared Chrome profile, work in background tabs, and continue through supported login and verification steps. Browse websites, fill forms, read pages and come back with evidence of what happened.

Less babysitting. More finished work.

The agent reaches a login page. A code lands in your inbox. A challenge appears after Submit. Then the conversation ends and nobody knows whether the application went through.

Ruin handles the browser work that keeps interrupting the task. Figure A shows the shared browser; B covers login and verification; C brings the result back to something you can check.

Core feature What you get
Your Chrome, ready to work Navigate, click, type, select and extract text across websites. Keep using the same profile, cookies and extensions between agent sessions.
An agent that stays out of your way Task tabs open in the background. Ruin keeps them interactive without switching your foreground app, and preserves unrelated tabs.
Log in once. Reuse it when you can. The local vault captures manual password logins and can fill a matching login later. Cookie and Web Storage snapshots help restore state after Chrome restarts. Sites still control expiry.
Keep moving through verification Supported flows handle fresh email codes, SMS codes and visual puzzles in the original task tab. Unresolved challenges return a pending state.
Less repeated account setup A local identity profile fills supported signup and contact forms. Optional Workday/iCIMS recovery handles recognized account errors and password-reset requests.
Proof you can come back to The Ashby watcher saves the site's submission confirmation by job URL. Look it up later—even after Chrome closes.

General browsing works through ruin_browser; the named sites below are examples of extra automation. The agent plans the task; Ruin runs its browser actions.

Attach a local file

Pass the exact tab URL as site and an absolute file_path to ruin_browser(operation="upload") or its alias set_files. As panel A shows, you can omit selector when exactly one file input exists across the page and its frames. For a form with multiple fields, select the intended input. Panel B shows the explicit drop-area route. Browser tests verify both paths, including a hidden input in a nested frame.

Local slider and rotation matching

Figure A shows deterministic local matching: ddddocr/OpenCV locate a horizontal gap, while NumPy compares RGB and angular gradients across a circular seam. Native coordinates are scaled to CSS pixels; rotation is calibrated against the actual rail. Feishu rotation markup takes priority over generic piece detection, including /campus/login. Figure B shows 15–25 ease-out moves with small Y jitter and a 150–250 ms hold before release. Flattened images require distinct matching outlines; ambiguous images and single images without a reference remain pending, without visual-model distance guesses. Figure C reflects a live NIO Feishu check: the SMS resend countdown appeared. This confirms the SMS request, not login or application submission.

Validation: 18 focused algorithm and Chrome fixture tests pass in the isolated patch snapshot. The existing working-tree retry change (8 attempts versus the older test expectation of 3) is outside this patch.

Three steps to your first browser task

1. Start the shared browser

You need Python 3.11+, Google Chrome and uv. This quickstart uses the current GitHub code so it includes the features described here.```sh
uvx --from 'ruin-everything @ git+https://github.com/yuki4266/ruin-everything.git@main' ruin browser


Ruin starts or reconnects to Chrome on `127.0.0.1:9222`. Its persistent profile is `~/.ruin-everything/canvas-browser`. Sign in to the sites you need **in that profile**; this does not import another Chrome profile's logins. Chrome and task tabs stay open when the command exits.

### 2. Connect your agent

Add this to your MCP client's server configuration, then restart the connection. The client must be able to run `uvx`.```json
{
  "mcpServers": {
    "ruin-everything": {
      "command": "uvx",
      "args": [
        "--isolated", "--refresh", "--from",
        "ruin-everything @ git+https://github.com/yuki4266/ruin-everything.git@main",
        "ruin-mcp"
      ],
      "env": { "RUIN_AUTO_UPDATE": "0" }
    }
  }
}

Prefer the published package? Replace the Git source with ruin-everything. Git updates arrive on the next server launch; PyPI features depend on the published version. Installation and update details.

3. Give it a task

> Use Ruin to open https://example.org and tell me the page title and visible links.

The entry point is ruin_browser(operation="navigate", url="https://example.org"). Follow-up calls use the returned current URL. No API key is required for Ruin's MCP server; your agent uses its own model connection. The optional standalone model-driven CLI needs the agent extras and OPENAI_API_KEY.

For account and signup helpers, install a persistent checkout and run ruin identity. Select the background editor tab to add your details and choose whether to enable automatic signup. ruin ats-setup separately opts into ATS password recovery. Setup and configuration.

Where it earns its keep

Cloudflare Turnstile continuation

Ruin locates the outer iframe through closed shadow roots, checks its visible state, and sends one bounded CDP mouse gesture. Ambiguous layouts, covering overlays and failed verification stay pending without replaying Submit. Screenshot fallback uses the local vision model. The existing Chrome profile is preserved; new launches include AutomationControlled suppression, with idempotent compatibility initialization on selected tabs and no fixed screen-coordinate offset.

Only genuine, unexpired cf_clearance cookies are backed up in the private ~/.ruin/cookies_snapshot.json namespace and restored for the same profile and browser user agent. Ordinary Turnstile often issues only a single-use token; token values are neither logged nor saved. A cleared widget is not an application receipt. See implementation, sources and live verification.

Optional Patchright support selects the same API and exception types for tasks and the persistent vault service. Install from this checkout with pip install -e ".[patchright]"; set RUIN_BROWSER_ENGINE=patchright, or save {"engine":"patchright"} to ~/.ruin-everything/browser-engine.json. Restart existing runtime clients and the vault when changing engines. Patchright owns initialization; Ruin skips its extra prototype patches. Connections have a bounded initialization timeout. Browser restarts remain explicit, preserving the original profile. See panel A below.

You're trying to… Ruin's extra help Verification so far
Get through a recruiting site's puzzle Handle the supported ZhongAn/Moka jigsaw or Feishu Jobs rotation slider. Live SMS countdowns observed on both; ZhongAn also returned its success page and application-history entry.
Finish a Greenhouse email check Match a fresh Outlook message to the employer and request, then enter its code in the job tab. Browser fixtures; live employer/Outlook compatibility remains unverified.
Recover a Workday or iCIMS account Use local identity details, supported signup/activation and optional bounded password recovery. Browser fixtures; live Workday dialog inspection. Production layouts vary.
Find out whether an Ashby application went through Watch for the official confirmation and save a receipt for the exact job. Local receipt and browser integration tests. A missing record means the outcome is unknown.
Get to a paper through Duke Follow supported institutional access on IEEE, ACM and Springer; continue filled Duke login and Duo device-trust screens. Live IEEE lookup, Duke redirect and PDF access; ACM/Springer covered by fixtures.

These are specific supported flows. Puzzle handling can require a local Ollama vision model; SMS reading requires macOS Messages access; mail flows need Outlook Web signed in. Phone approvals and unsupported checks can still need you. Workflow requirements and commands.

For authorized form submissions, use ruin_action with the exact current URL and visible button label. It checks for verification after the click. A cleared challenge is a step forward; the site's receipt is the evidence of submission.

Follow an ordered GeeTest icon prompt

Ruin detects the visible icon panel, uses ddddocr to locate candidate boxes, and asks the local Ollama vision model to match numbered crops to the prompt from left to right. It maps bitmap centers to CSS viewport coordinates, clicks in order, and presses OK automatically. Failed attempts refresh only the widget, with at most three attempts total. The Bilibili icon challenge was cleared live in one attempt; login and application submission are separate checks. Strong rotated-template matches cross-check the order; otherwise each target symbol is matched separately. Browser tests cover DPR, iframe scaling, stale images and bounded retries.

Enter a code across several boxes

Panel A shows how Ruin matches an incoming SMS or supported Outlook message. Panel B shows how it pastes the code into a split form and fills any remaining cells. The Ctrip phone check was verified live. Both input paths were tested with react-otp-input 3.1.1. Single code fields can also be identified by their visible placeholder or accessible name; ambiguous fields are left untouched. Bilibili SMS messages are matched to the exact login host and request time.

Six tools. One shared browser.

Tool Use it for
ruin_browser List tabs, navigate, inspect, read, click, fill, press keys and select options.
ruin_action Submit an authorized form and handle supported follow-up verification; omit the button to resume.
ruin_auto_register Continue supported signup, account or SMS-login steps using the local identity profile.
inspect_application_challenge Identify a Greenhouse/Ashby challenge or locate the supported application tab.
get_ashby_receipt Look up saved submission evidence without opening Chrome.
watch_ashby_receipt Watch the original Ashby tab for confirmation and save it.

Agent routing instructions · CLI and detailed behavior · Trusted-login configuration

Your machine. Your browser state.

Browser execution and credential handling run locally. The automatic login vault stores plaintext passwords in ~/.ruin/auth_vault.json with owner-only permissions (0600); they stay out of agent tool results. Cookies and Web Storage are also backed up locally. Set RUIN_LOCAL_AUTH_VAULT=0 to disable automatic listener startup for that client. Vault behavior and storage limits.

Your connected agent may send page text to its model provider. Site sessions still expire, and some verification still needs human input. Ruin preserves browser state and resumes supported steps; it cannot promise permanent login or success on every site.

What has actually been checked

Current check: Turnstile continuation and optional Patchright support are now consolidated on main (Turnstile figure, panels A–D). All 133 regression tests passed in one run with the Playwright engine; source/wheel builds and strict Twine metadata checks also passed. The checks figure above distinguishes local fixtures from live site evidence. Prior live observations and their limits remain in the runtime reference.

The failed CI run stopped on Ubuntu mirror DNS errors while fetching extra font packages. Hosted Ubuntu already supplied the Chromium shared libraries. CI now installs Chromium directly and validates the runner through the browser regression suite (figure A–C). Browser downloads get at most three attempts, with 15s and 30s waits and a 30-minute step timeout. Successful tests and packaging checks enable PyPI publishing. CI results.

Local tests cover shared sessions, background tabs, login capture and restoration, tenant boundaries, verification routing and receipt storage. The table above distinguishes live observations from simulated browser tests. See the reference for the conditions and limits of each check.

To run the regression suite from a development checkout:sh python -m pip install -e '.[agent]' python -m playwright install chromium RUIN_AUTO_UPDATE=0 RUIN_RULE_UPDATES=0 python -m unittest discover -s tests -v

Put it to work. Tell us where it gets stuck.

Report a bug · Share a workflow · Suggest a feature

If Ruin saved you another round of browser babysitting, give it a star.

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Ruin Everything 怎么安装?

在客户端按下面的安装命令接入:uvx --from 'ruin-everything @ git+https://github.com/yuki4266/ruin-everything.git@main' ruin browser。装完在客户端配置里引用同一个包或端点。

Ruin Everything 怎么在 Claude、Cursor、VS Code 里配置?

它的传输方式是HTTP · 远程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Ruin Everything 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Ruin Everything 需要 API Key 吗?

资料标注需要会话登录认证,使用前准备好对应凭据,并按最小权限配置。

Ruin Everything 支持哪些客户端?

资料标注支持 Claude、Cursor、VS Code、Gemini CLI、Codex、Zed;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Ruin Everything 能提供哪些能力?

资料显示能力标签包括 开发。具体工具清单以官方文档为准。

Ruin Everything 需要什么运行环境?

属于远程端点,不需要在本地装运行时。

Ruin Everything 是本地还是远程?

它是远程端点,连服务地址即可。传输方式为HTTP · 远程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-11-20。