大量的工具说明书挤占了模型的上下文。根据Anthropic的计算，大概50 个工具的定义就会吃掉约 20,000 Tokens。结果Agent 的注意力全放在记住工具名上了，其他的执行、推理严重受损。

What if the way AI agents interact with tools and resources could be as seamless as browsing the web? Imagine a world where developers no longer wrestle with custom-built adapters or fragmented ...

11 月 25 日，Anthropic 官方博客更新了一篇庆祝帖，宣布 MCP 正式满一周岁，配合发布的还有一份新版规范。 官方给出的数据看起来不错：MCP Registry 目前收录了近 2000 个 Server，比 9 月份刚上线时增长了 407%。OpenAI 在 3 月宣布全面支持 MCP，Google、AWS、HuggingFace 等都 ...

These milestones are meaningful, but they don’t tell the full story: MCP succeeded because it removed the friction inherent ...

MPC replaces bespoke per-resource proprietary connections and has become immensely popular across the AI spectrum ...

01. builder 比 user 还多，MCP 仅是「旧瓶装新酒」？ 一年过去，社区对于 MCP 的定位仍有争议？平均 25 个用户对应 1 个开发者，MCP 目前更多是开发者自娱自乐的产物？... MCP 大规模落地还得看下一个「微信小程序」入口的出现？... builder 比 user 还多，MCP 仅是「旧 ...

The Model Context Protocol (mCP) is reshaping how artificial intelligence (AI) systems interact with data, tools, and environments. Developed as an open source standard by Anthropic, mCP simplifies ...

MCP certification is a hot topic in AI hiring. Here’s what courses are available now, when official certification might arrive, and why it may not solve every problem. The Model Context Protocol (MCP) ...

Two critical remote code execution vulnerabilities in the Model Context Protocol (MCP) ecosystem have laid bare the hidden risks lurking in what's quickly becoming AI's new backbone infrastructure.