Pancake 推出 MCP 服务器，让编程智能体自行完成获客
Pancake 做了个 MCP 服务器，让 Claude、Codex 这类编程智能体写完代码后还能自己找客户，聚合 40 多家数据源一条流程跑完。
Pancake 发布了一个 MCP 服务器，可接入 Claude 或 Codex 等编程智能体。它聚合 40 多家数据提供商做线索搜索和画像补全，并支持用用户自己的账号发送消息。产品还会根据用户网站生成一套 GTM 上下文，让智能体把“查找正在招聘市场负责人的获投初创公司”这类任务作为工具调用执行，返回草拟好的外联消息。此前工程师需要手动接入 5 家厂商的 API 才能实现同样流程。
Coding agents can now ship a product end to end, and then they hit a wall: nobody gave them tools for finding customers.
Pancake is that tool surface. One MCP server gives Claude or Codex lead search across 40+ data providers, enrichment, and sending from your own account, plus a GTM brain built from your website.
So the same agent loop that wrote the code can run “find funded startups hiring a marketing lead” as a tool call and come back with drafted messages.
GTM ("Go to Market") becomes a problem of data pipelines, context management and sending infrastructure, with the agent as the caller. Until now that meant an engineer wiring five vendor APIs into an agent by hand. Here the wiring sits behind one server.
