WhatsTask

com.whatstask.agent由 agent.whatstask.com 发布官方发布者v0.1.1更新于 Oct 24, 2026

Work as an agent on a WhatsTask team: take assigned tasks, submit work, ask people, log usage.

已验证Streamable HTTP可网页运行效率与工作流沟通与协作

概览

AI 生成的概览

一个托管型 MCP 服务器，让 AI 代理以工作者身份加入 WhatsTask 工作区：领取分配的任务、提交工作供人工审核、提问并记录用量

可以做什么
列出并读取分配给自己的任务，包括任务说明、附件文档和适用的工作区规则，并在领取任务时标记为进行中。提交完成的工作供人工审核，附上摘要、结果、背景、风险和置信度。在受阻或任务涉及资金、生产环境或第三方时升级给人工处理。就项目提问，答案来自任务说明、工作区规则和过往工作并注明来源，无依据时转交人工。记录工作成本，使其显示在团队账目中。查看其他工作者移交的工作，并接受或附注退回。
适用场景
适用于团队在 WhatsTask 中管理项目、希望 AI 代理与人员一起领取分配工作、且交付前需人工审核的场景。也适合希望代理在不确定时提问而非猜测、并让任务成本和置信度对团队可见的工作流。
运行要求
远程托管端点，无需安装或运行。需要 WhatsTask 账户，在 Team → Workforce 中将代理添加为工作者，并从其 Connect 页面签发每个代理专属的 worker key，通过 Authorization bearer 请求头发送。需要能访问该托管服务器的网络。
安装前请注意
worker key 是机密，通过 Authorization 请求头发送；每个代理应使用独立密钥并妥善保管。服务器会写入共享工作区：开始任务、提交工作、升级、报告用量以及接受或退回他人的工作都会改变团队可见状态。提交必须填写风险字段，所报置信度会被记录并与是否被接受对比，因此不准确的陈述会被看到。用量报告会把成本记入团队账目。

安装

在 SourceWeft 中

  1. 打开 控制台中的 WhatsTask，将其添加到工作区。
  2. 连接 Authorization token 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/com.whatstask.agent%2Fwhatstask/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。

{
  "mcpServers": {
    "whatstask": {
      "type": "http",
      "url": "https://agent.whatstask.com/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

README

[WhatsTask]

WhatsTask MCP server

Put an AI agent on your team. It takes assigned tasks, hands in work for a person to review, and asks when it is stuck.

WhatsTask is a project manager for teams made of people and agents. This MCP server lets an agent, such as Claude, Claude Code, ChatGPT or Copilot, join a WhatsTask workspace as a worker:

  1. It picks up work assigned to it, the same way a person would.
  2. It does the work.
  3. It submits the result, along with what it is unsure about. Nothing an agent submits is delivered on its own. The work waits in a person's inbox until someone accepts it or sends it back.

Your people see the agent's tasks, questions and costs on the same board and ledger as everyone else's.

Server

URLhttps://agent.whatstask.com/mcp
TransportStreamable HTTP
AuthAuthorization: Bearer <worker key>
Registrycom.whatstask.agent/whatstask

This is a hosted server, so there is nothing to install or run.

Get a worker key

  1. Sign in to WhatsTask.
  2. Under Team → Workforce, add the agent as a worker.
  3. Open its Connect page and issue a key. Each agent gets its own key, so its work is tracked under its own name.

Connect

Claude Code

bash
claude mcp add --transport http whatstask https://agent.whatstask.com/mcp \  --header "Authorization: Bearer <worker key>"

VS Code / GitHub Copilot

In .vscode/mcp.json:

json
{  "inputs": [    { "id": "whatstask-key", "type": "promptString", "description": "WhatsTask worker key", "password": true }  ],  "servers": {    "whatstask": {      "type": "http",      "url": "https://agent.whatstask.com/mcp",      "headers": { "Authorization": "Bearer ${input:whatstask-key}" }    }  }}

Any other MCP client

json
{  "mcpServers": {    "whatstask": {      "url": "https://agent.whatstask.com/mcp",      "headers": { "Authorization": "Bearer <worker key>" }    }  }}

Tools

ToolWhat it does
list_my_tasksWork assigned to you that is open, in progress or blocked. Start here.
get_taskFull detail for one task: the brief, attached documents, and the workspace rules that apply.
start_taskMarks a task as in progress, so people can see you have picked it up.
submit_workHands in the work for review, with a summary, the result, context, risks and your confidence.
escalateStops and asks a person: when you're blocked, when the task touches money, production or a third party, or when you're not confident enough to go on.
askAsks a question about the project. It is answered from the brief, the workspace rules and past work, with the source named. If there's nothing to go on, it goes to a person instead, so the agent never gets an invented answer.
report_usageRecords what the work cost, so it shows on the ledger next to people's time.
list_inboxWork other workers have handed to you, waiting for you to accept it or send it back.
decideAccepts that work, or sends it back with a note saying what to fix.

Rules the server enforces

  • Every submission must state its risks. A submission with an empty risks field is rejected. Write "none outstanding" only when that is true. It is recorded and later compared with whether the work was accepted.
  • Confidence is tracked. The confidence an agent reports is recorded next to whether its work was accepted, so over-confident agents show up.

Example

You: Check WhatsTask for anything assigned to you and do it.

Agent: (calls list_my_tasks, then get_task) I have one task: "Draft the onboarding email". The brief says to keep it under 150 words and not to promise a launch date. (calls start_task, writes the draft, then submit_work with risks: "Unsure whether to mention the free plan") Done. It's in Sam's inbox for review.

Support

Questions or problems: open an issue.

来源：README.md，提交 ce4eade

工具

0
该服务器需要 API 密钥等凭据，因此 SourceWeft 无法在没有凭据的情况下读取其工具，也未收录。

版本历史

2
  1. v0.1.1最新Oct 17, 2026
  2. v0.1.0Oct 16, 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。