Pi.dev：从拒绝到支持MCP

Pi开发平台改变了此前的决策，宣布在升级版本中将MCP（Model Context Protocol）集成为核心功能1。这一转变源于MCP自身的显著改进，同时Pi团队认识到为MCP所做的相关技术调整也能为Jev等其他平台工具提供通用基础设施1。

为了解决MCP的工具组合能力问题，Pi引入了Codemode沙箱机制1。Codemode是一个运行在主程序端的JavaScript沙箱，用于协调和组织工具调用1。在这一机制下，工具可以返回结构化数据，并能通过文档和描述实现智能发现，其设计理念接近OpenAPI标准1。Pi团队指出，现代MCP相比过去已有改进，但MCP服务器和协议模式本身仍存在进一步改进的空间1。

Pi's development platform has announced a significant reversal of its previous stance by integrating the Model Context Protocol (MCP) as a core feature.1 The decision comes after the platform explicitly stated it would not support MCP on pi.dev, but the team now recognizes improvements in MCP itself as well as the broader benefits such integration could provide.1

The primary motivation behind adding MCP support stems from enhancements to the protocol itself and the recognition that infrastructure changes required for MCP would also serve other Pi platform tools more effectively.1 Specifically, the modifications would create a shared foundation beneficial to features like Jev.1 To strengthen MCP's tool coordination capabilities, Pi has introduced Codemode, a JavaScript sandbox running on the harness side that organizes and coordinates tool calls.1 This mechanism enables tools to return structured data and support intelligent discovery, aligning with design principles similar to OpenAPI.1

While the Pi team acknowledges that contemporary MCP represents marked improvement over earlier iterations, they recognize that MCP servers and patterns still have room for enhancement.1