AI UNDERDOGSDAILY PICK
AI UNDERDOGS
MCP 挂了，你的 Agent 不会说
Your MCP broke. Agent won't tell you
Spanly
你的 MCP 服务器挂了
AI Agent 不会报错
它只会悄悄给你一个错误答案
或者直接跳过工具调用——你以为一切正常
Your MCP server breaks mid-conversation. The agent doesn't
error out. It silently drops tool calls or
hallucinates results — and you'd never know
MCP 层的盲区
The blind spot in your stack
Datadog、Sentry
New Relic——它们能监控你的 API 和后端
但 MCP 工具调用这一层是完全的盲区
Spanly 不替代它们
而是坐在旁边，只盯这一个没人管的层
Datadog, Sentry, New Relic — they watch your
APIs and backend, but the MCP tool-call layer
is invisible to all of them. Spanly sits
alongside them, scoping only the gap your existing
stack can't see
★ SIGNAL 1
一条 curl 诊断
One curl, 30-second diagnosis
一行 curl 命令跑一遍
30 秒出结果——133 多个扫描器检查你的 MCP 配置
输出一个字母评级和按优先级排好的修复清单
零代码改动
然后它持续监控生产环境每次工具调用
自动写修复
Run one curl command — 30 seconds, 133-plus
scanners grade your MCP setup with a letter
score and a ranked fix list. Zero code
changes. Then it monitors every production tool call
and auto-writes the patches
★ SIGNAL 2
一个人撑起整个垂直
One person, full vertical
SDK、命令行工具
GitHub 上的开源扫描器
加上托管平台——全部出自同一个作者
从诊断到监控到自动修复
一个人把整个垂直线做完了
这才是 indie hacker 该有的打法
SDK, CLI, open-source scanners on GitHub, and a
hosted platform — all shipped by one author
Diagnosis, monitoring, auto-fixes — one person built the
entire vertical. Classic indie hacker move
作者自己说的：「Agents won't say your MCP broke. Spanly will.」——很直接
没有花哨话
你现有的监控栈看不到 MCP 层
Agent 本身也不会承认出了问题
所以你需要一个专门盯这个的东西
The author puts it bluntly: "Agents won't say
your MCP broke. Spanly will." Your monitoring stack
can't see the MCP layer, and the agent
itself won't admit something broke — so you
need something that only watches this
AI UNDERDOGS
MCP 可观测性，一个人做出来的
MCP observability, indie-built
Spanly
关注 · 每天发现更多 AI 神作
spanly.com