惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

博
博客园 - 叶小钗
J
Java Code Geeks
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
阮一峰的网络日志
阮一峰的网络日志
爱范儿
爱范儿
量
量子位
N
Netflix TechBlog - Medium
博
博客园 - 聂微东
博
博客园 - Franky
aimingoo的专栏
aimingoo的专栏
The Cloudflare Blog
T
The Blog of Author Tim Ferriss
MyScale Blog
MyScale Blog
Google DeepMind News
Google DeepMind News
小众软件
小众软件
博
博客园 - 三生石上(FineUI控件)
C
Check Point Blog
钛媒体：引领未来商业与生活新知
钛媒体：引领未来商业与生活新知
B
Blog
Engineering at Meta
Engineering at Meta
Microsoft Azure Blog
Microsoft Azure Blog
博
博客园_首页
H
Hackread – Cybersecurity News, Data Breaches, AI and More
腾
腾讯CDC

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
MCP Sentinel v1.0 Is Out: A Lockfile for MCP Tool Schemas
Wannavf · 2026-05-09 · via DEV Community

A few days ago I posted about a problem I hit with MCP tool schema drift.

The short version:

An MCP server changed a parameter from location to city.

My agent kept sending location.

Nothing warned me before runtime.

That first version of MCP Sentinel was just the basic idea:

sentinel init
sentinel snapshot
sentinel check
sentinel diff

Enter fullscreen mode Exit fullscreen mode

Since then, I shipped MCP Sentinel v1.0.0.

GitHub:

https://github.com/Wannavf/mcp-sentinel

Enter fullscreen mode Exit fullscreen mode

npm:

npm install -g @wannavf/mcp-sentinel

Enter fullscreen mode Exit fullscreen mode

The framing changed a bit

The original problem was annoying because it broke my tools silently.

But after feedback, I think the bigger point is this:

MCP tool schemas are contracts.

If a tool changes its parameters, required fields, enum values, return shape, or descriptions, that is not just a small implementation detail.

It changes what an agent can ask for.

It can also change when the agent decides to use that tool.

That matters a lot for MCP servers around:

databases
filesystems
cloud infrastructure
admin tools
internal APIs

Enter fullscreen mode Exit fullscreen mode

A broken weather tool is annoying.

A silently changed database or infrastructure tool can be much worse.

What v1.0 adds

The first version was basically:

snapshot
check
diff

Enter fullscreen mode Exit fullscreen mode

v1.0 now has a fuller workflow.

sentinel init
sentinel discover --write
sentinel doctor
sentinel snapshot
sentinel check
sentinel diff

Enter fullscreen mode Exit fullscreen mode

Discovery

You can now ask Sentinel to find MCP servers from common config locations:

sentinel discover
sentinel discover --write

Enter fullscreen mode Exit fullscreen mode

It can look for MCP-shaped configs and help import servers into sentinel.config.json.

This makes onboarding less manual.

Doctor

There is now a setup check:

sentinel doctor

Enter fullscreen mode Exit fullscreen mode

It checks whether Sentinel can see your config, whether servers are configured correctly, and whether the lockfile exists.

Instead of guessing why something does not work, you get a quick sanity check.

Dashboard

There is also a terminal dashboard:

sentinel dashboard

Enter fullscreen mode Exit fullscreen mode

or:

sentinel db

Enter fullscreen mode Exit fullscreen mode

It lets you inspect configured servers, snapshots, checks, activity, and drift locally.

It is not the main feature, but it is useful when testing MCP servers during development.

CI

The main use case is still CI.

sentinel check

Enter fullscreen mode Exit fullscreen mode

If a server changed its tool contract, CI can fail before the agent discovers the break at runtime.

There is also a GitHub Action:

name: MCP Schema Check
on: [pull_request]

jobs:
  drift:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: Wannavf/mcp-sentinel@main
        with:
          fail-on: MAJOR

Enter fullscreen mode Exit fullscreen mode

Reports

Sentinel can output diffs in multiple formats:

sentinel diff --format json
sentinel diff --format markdown
sentinel diff --format sarif

Enter fullscreen mode Exit fullscreen mode

SARIF can be uploaded to GitHub Code Scanning.

Transports

v1.0 supports:

stdio
Streamable HTTP
SSE

Enter fullscreen mode Exit fullscreen mode

Example stdio config:

{
  "compatibility": "BACKWARD",
  "failOn": "MAJOR",
  "servers": {
    "filesystem": {
      "transport": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}

Enter fullscreen mode Exit fullscreen mode

Example HTTP config:

{
  "servers": {
    "remote": {
      "transport": "http",
      "url": "http://localhost:3000/mcp"
    }
  }
}

Enter fullscreen mode Exit fullscreen mode

What it catches

Sentinel classifies changes as MAJOR, MINOR, or PATCH.

Examples of MAJOR changes:

tool removed
required parameter removed
parameter type changed
optional parameter made required
constraint tightened
enum value removed
output field removed

Enter fullscreen mode Exit fullscreen mode

Examples of safer changes:

optional parameter added
tool added
enum value added
description updated

Enter fullscreen mode Exit fullscreen mode

The goal is not to block every change.

The goal is to make tool contract changes explicit.

Why I think this matters

Without a lockfile:

MCP server changes
agent breaks later
you debug at runtime

Enter fullscreen mode Exit fullscreen mode

With a lockfile:

MCP server changes
CI detects drift
you review the contract change
you decide whether to accept it

Enter fullscreen mode Exit fullscreen mode

That feels like a missing piece in MCP tooling.

Package managers have lockfiles.

APIs have contracts.

MCP tool schemas should have a review point too.

Try it 

npm install -g @wannavf/mcp-sentinel

Enter fullscreen mode Exit fullscreen mode

or:

npx -y @wannavf/mcp-sentinel --help

Enter fullscreen mode Exit fullscreen mode

GitHub:

https://github.com/Wannavf/mcp-sentinel

Enter fullscreen mode Exit fullscreen mode

npm:

https://www.npmjs.com/package/@wannavf/mcp-sentinel

Enter fullscreen mode Exit fullscreen mode

I would especially like feedback from people using MCP with database, filesystem, infrastructure, or internal admin tools.

此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。