Armature（YC P26）创业公司推出了针对MCP的产品分析和评估工具，使开发者能够通过仅需3行代码的SDK集成来追踪Agent会话、用户意图和使用模式[1]。该工具支持Typescript、Python和Go三种编程语言[1]，自助设置耗时少于5分钟，并提供免费套餐[1]。
该产品通过客户端隐私保护、自动用例聚类和问题识别等功能，帮助MCP开发者理解用户如何使用其工具及遇到的问题[1]。性能测试结果显示，集成该工具后的成功率为89.17%，而无instrumentation时的成功率为89.15%，基于870次运行的测试数据[1]。一个营销自动化平台客户通过该工具的分析发现了用户痛点并进行了修复，随后的评估测试又识别出小模型存在hallucination风险[1]。
创始人Louis曾开发过暴露给数百万用户的MCPs，而Theo为Palantir前瞻部门工程师和Datadog衍生公司创始工程师[1]。该团队计划进一步集成自动化修复和跨模型测试能力[1]。
Armature, a Y Combinator P26 startup, has introduced a product analytics and evaluation platform designed for developers building Model Context Protocol (MCP) applications.[1] The tool enables developers to integrate an SDK with just three lines of code to track agent sessions, user intent, and usage patterns.[1] The platform assists MCP developers in understanding how users interact with their tools and identifying issues they encounter.[1]
The SDK is available in three programming languages: TypeScript, Python, and Go.[1] Setup can be completed in less than five minutes, and the company offers a free tier for users.[1] Performance testing showed a 89.17% success rate compared to 89.15% without instrumentation across 870 test runs, demonstrating minimal overhead.[1] The platform incorporates client-side privacy protection and automatic use-case clustering for problem identification.[1] Future roadmap plans include automated remediation integration and cross-model testing capabilities.[1]
A marketing automation platform used the analytics to discover user pain points and implement fixes, while subsequent evaluation testing uncovered hallucination risks in smaller language models.[1] The company's founders bring relevant expertise: Louis has built MCPs exposed to millions of users, while Theo previously worked as a forward deployment engineer at Palantir and co-founded a Datadog spinoff company.[1]