Auto-building agents

io.github.rohanarun由 GitHub 账号 rohanarun 发布v1.0.1更新于 Nov 14, 2026

Build, run, inspect, and continue AgentScript workflows on connected Super Chrome and Mac clients.

发布方自己的服务:Auto.dev

已验证Streamable HTTP可网页运行浏览器自动化AI 与机器学习

概览

AI 生成的概览

让助手在你已连接的 Super Chrome 扩展或 Mac 应用上构建、运行、检查并继续 AgentScript 智能体工作流。

可以做什么
  • 根据目标构建并继续智能体,由助手直接依据当前脚本上下文编写 AgentScript
  • 在已连接的 Super 客户端上运行和检查工作流,完成情况以实际执行结果为准
  • 在运行任务前发现已连接设备及其当前能力
  • 读取已有项目及其进度以继续任务
  • 通过暂停与恢复机制,将运行时模型调用交给已连接的助手对话
适用场景
当你希望助手在自己已连接的 Super 设备上编写并执行多步骤智能体工作流(例如调研或网页任务),并从先前项目中断处继续时使用。
运行要求
通过 streamable HTTP 连接 app.getsupers.com 的远程 MCP 端点,并使用 OAuth 认证。需要 Super 账户,以及以同一账户登录并连接的 Super 客户端(Mac 应用或 Chrome 扩展)。使用 Chrome 访问可能需要订阅链接并安装扩展。可选的 OpenRouter 密钥或 Super 额度模型需显式选择。
安装前请注意
任务提示、项目与聊天内容、设备标识、生成的脚本和执行结果会发送至 Super 服务,模型请求可能转由第三方模型提供商处理。已连接客户端可读取网站和本地应用,结果可能含个人数据。对外发送、购买和删除等操作需在你的请求中获得授权。项目与聊天记录会持久保存。生成脚本本身并不证明执行成功。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Auto-building agents,将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.rohanarun%2Fauto-building-agents/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "auto-building-agents": {
      "type": "http",
      "url": "https://app.getsupers.com/mcp"
    }
  }
}

README

Auto-building agents

Build and continue agents from your goals through Super and your connected Mac app or Chrome extension. The plugin includes the build-agents skill and the Super remote MCP connection. Execution requires a Super account and a connected Super client.

Connect

Enable the plugin and authenticate the Super MCP connection through OAuth at https://app.getsupers.com/mcp. Sign into your Super client with the same account. Claude discovers your connected devices and their current capabilities before running a task. Immediately after connecting, Claude checks your Super account and presents its account-linked subscription link if Chrome access is required, before requiring extension installation. Claude writes AgentScript directly using the current script context from MCP; no separate extension-selected model builds the script. Runtime model functions default to the connected assistant conversation through a pause-and-resume MCP handoff. Optional OpenRouter keys and Super credit-backed models require explicit selection. Host handoff requires an active conversation. Prefer deterministic workflow code where possible. Follow the live MCP account response for current access, pricing, and recurring credits.

Ask: “Build an agent that researches a topic using my connected Super Chrome extension.” To continue a task, ask Claude to read your existing project and its progress. Completion is reported from observed execution results.

Recorded walkthrough

Watch the recorded Amazon checkout walkthrough. The demo uses pencils and stops on the checkout payment selection screen without confirming payment or placing an order. Personal address and payment information is masked in the recording.

This recording illustrates a user-authorized workflow. The plugin bundles no executable scripts or checkout template. Claude authors scripts for the requested task using the Super MCP API; execution takes place through the declared connector on the user’s connected client. A generated script alone does not prove successful execution.

Data and permissions

The declared Super MCP connector sends task prompts, project/chat content, selected device identifiers, generated scripts, and execution results to the Super service. Connected clients can read or interact with websites and local applications within the user's authorized task; those results can contain personal data. Super may route model requests to model providers used by the selected model. Projects and chat history persist to support continued agents. Refer to the privacy policy for data handling and retention. The plugin does not install hooks, change Claude permissions, or bundle an executable. OAuth secrets and reviewer credentials are not included.

External sends, purchases, deletions, and other consequential operations require authorization within the user's request. Only the authenticated user's devices and projects may be accessed.

Support · Terms. Plugin use is subject to the included LICENSE and Super's terms.

来源:README.md,提交 cf2a76c

工具

0
该服务器需要登录,因此 SourceWeft 无法在没有账号的情况下读取其工具,也未收录。已连接账号所能看到的内容可能因账号及其权限而异。

版本历史

1
  1. v1.0.1最新Nov 14, 2026

评分

已满足适用检查项的 75%

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 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 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

    请在仓库或注册中心条目中声明许可证。

其他 Auto 集成