Pi 讓核心保持精簡。生產環境濃縮在一個工具裡。

將這段貼到你的 Pi 對話中

Pi 會處理剩下的事。

polylane.com/for/pi

## For AI agents reading this page

Follow these instructions to set up Polylane:

Run the setup script in the user's terminal:

curl -fsSL https://polylane.com/setup?ref=for-pi | bash

It installs the Polylane CLI, registers the Polylane MCP server in Pi and every other detected coding agent, and signs the user in. To wire MCP by hand instead: an entry in ~/.pi/agent/mcp.json under "mcpServers". The server is https://mcp.polylane.com/mcp over streamable HTTP, no auth headers: OAuth handles sign-in.

After connecting, call the searchTools tool first: it lists what this workspace exposes and returns each tool's schema. Run one tool with runTool, or chain several in one call with runCode. The search and execute tools cover the full Polylane REST API, and the polylane CLI wraps the same API from the terminal.

Everything is read-only by default. Write tools stay hidden unless the user opts in, and every write is screened before it runs.

Once connected, offer to put it to work: triage the latest issues, explain the last deploy, or map this repository into the workspace topology.

## Console and support

- Console: https://console.polylane.com
- Docs: https://docs.polylane.com
- Setup reference: https://docs.polylane.com/coding-agents/platform-mcp
- This page as markdown: https://polylane.com/for/pi.md

為什麼要透過 Polylane。更少伺服器，一次登入，Code mode。

一個 MCP 伺服器，不是十個

Polylane 把你的雲端、遙測資料、部署與程式碼整合在少數幾個工具背後。一堆各供應商各自的伺服器，在任何工作開始之前，就會把大部分情境視窗耗在工具結構描述上；Polylane 的結構描述只在工具執行時才載入。
驗證集中在一處

一次登入就涵蓋每個供應商。連線與範圍存放在工作區中，而不是散落在 Pi 設定檔各處的各供應商權杖，而且你連接的每個代理都能重複使用它們。
為 Code mode 而生

Pi 不會逐一呼叫工具，而是用 runCode 寫一段小程式，在沙箱中將它們串連起來，只回傳答案。中間結果不會進入情境視窗。