把 ZooWork 連接到 Busabase
An agent delivery platform: build agents in chat with Agent Builder, or run them through the Managed Agents API in an isolated cloud sandbox with shell, files, web tools, Skills and remote MCP servers.
把這段貼進 ZooWork
請閱讀 https://busabase.com/SETUP_SKILL.md,並幫我設定好 Busabase。請照著文件一路做到完成,最後確認連線是否正常運作。

把這一句指令交給你的 Agent
你得到的不只是一個連接器:busabase 技能還帶著完整的變更請求工作流程、CLI 與完整 API —— 之後每一次工作階段都在,不必再重新貼一次。
會先問你
Busabase Cloud 還是個人桌面版 —— 在動手之前,由你決定。
完成連線
幫你登入並自行儲存憑證,不需要手動複製 API 金鑰。
建立一個資料表
建立並填入你的第一個資料表,讓工作區一開始就不是空白畫面。
安裝技能
busabase 與 busabase-app-creator 會永久安裝,之後每一次工作階段都能直接使用。
此 Agent 的注意事項
Sign-in spans two turns: the agent stops after posting the approval link, so send a follow-up message in the same Session once you've approved it. Don't deny exec or web_fetch in tool_policy, or the agent can't read the setup document or run busabase-cli.
其他連接方式
Busabase 無法以 MCP 連接器的方式接入這個智能體,沒有連接器可以加入。可以改用以下方式。
一覽表
- 廠商
- Serendipity One
- 連接方式
- 外掛/技能 · busabase-cli · REST API
- 傳輸方式
- Streamable HTTP
- 驗證方式
- 無
npm install @zoowork-ai/sdk # TypeScript SDK
python -m pip install zoowork # Python SDKZooWork's MCP integration cannot reach Busabase: an Agent's mcp servers must be public endpoints that accept unauthenticated requests, and the platform attaches no stored credentials, OAuth, or custom headers. Connect through the Agent's own tools instead.
-
Create and start an Agent by following the Quickstart. Keep
execandweb_fetchavailable — they are on the default tool surface unless yourtool_policydenies them. The default cloud sandbox ships Node.js 24 and has unrestricted outbound network access. -
In a Session, send the agent this message:
Read https://busabase.com/SETUP_SKILL.md and set up Busabase for me. Follow it through to the end, then confirm the connection works.
-
The agent runs
busabase-cli login --no-waitand replies with an approval link. Approve it in your browser, then send another message in the same Session so the agent can finish signing in. -
Ask the agent to run
busabase-cli whoamito confirm the connected Space.
For a production integration where your backend already holds a Busabase API key, ZooWork recommends an application-executed custom tool for authenticated services: your application calls the Busabase REST API and returns the result to the agent.
已知的注意事項
MCP is not a route in: ZooWork only connects to public MCP endpoints that accept unauthenticated requests, with no OAuth, stored credentials, or header configuration — Busabase's MCP server requires sign-in. Managed Agents is a Developer Preview, so the API may change. busabase-cli stores its credential in the home directory, while only /workspace is documented as shared across an Agent's sessions, so a new sandbox instance may need to sign in again (not yet tested by us). Sessions on the same Agent share its workspace and therefore the Busabase sign-in; use an Agent per user for separate identities.
能力比較
與目錄中其他已收錄的 Agent,並列比較連線能力。
| 能力 | |||
|---|---|---|---|
| 類型 | Agent 平台 | 終端機 Agent | AI 編輯器 |
| 推薦路徑 | Agent Skill | Agent Skill | Agent Skill |
| 連接方式 | 外掛/技能 · busabase-cli · REST API | MCP + OAuth · 外掛/技能 | MCP + OAuth · 外掛/技能 |
| 傳輸方式 | Streamable HTTP | Streamable HTTP | Streamable HTTP |
| 驗證方式 | 無 | OAuth(動態註冊) | OAuth(動態註冊) |
把這段貼進 ZooWork
請閱讀 https://busabase.com/SETUP_SKILL.md,並幫我設定好 Busabase。請照著文件一路做到完成,最後確認連線是否正常運作。