More screenshots
Works with the AI Agents you already use
AgentBreakWatch Stack and MCP Drift Auditor
Audit AI agent manifests and .mcp.json files against 258 tracked Agent SDKs and MCP servers. Flags archived @modelcontextprotocol/server-* packages, unpinned @latest entries, and high schema-drift risk dependencies with exact pin commands.
Free
AgentBreakWatch Stack and MCP Drift Auditor
Example session with this skill installed
Audit our repository's package.json and .cursor/mcp.json for MCP schema drift, archived servers, and breaking Agent SDK updates.
- Read your context and instructions
- Compiled the agentbreakwatch stack
AgentBreakWatch Stack Health Summary
- Manifests Scanned:
package.json,.cursor/mcp.json - Tracked Packages Matched: 6 (3 Agent SDKs, 3 MCP Servers)
Archived Reference Servers Found: 2 (@modelcontextprotocol/server-puppeteer, @modelcontextprotocol/server-postgres)
- Unpinned
@latestMCP Entries: 2 - Overall Stack Drift Risk: CRITICAL
Prioritized Remediation Table
| Package / Server | Current Config | Live Version | 90d Releases | Drift Risk | Issue Summary | Exact Fix / Pin Command |
|---|---|---|---|---|---|---|
@modelcontextprotocol/server-puppeteer | npx -y ...@latest | v2025.5.12 (Archived) | 0 | HIGH | Archived on npm; replace with Playwright MCP accessibility snapshot server | npx -y @playwright/mcp@0.0.83 --headless |
@modelcontextprotocol/server-postgres | npx -y ... | v0.6.2 (Frozen) | 0 | HIGH | Frozen reference server lacking parameterized query guards and Streamable HTTP | npx -y @bytebase/dbhub --transport stdio --dsn ${DATABASE_URL} |
@modelcontextprotocol/sdk | ^1.20.0 | v1.32.2 (v2 at 2.4.0) | 5 | HIGH | v2 split into @modelcontextprotocol/server & @modelcontextprotocol/client (v2.4.0) | npm install --save-exact @modelcontextprotocol/server@2.4.0 @modelcontextprotocol/client@2.4.0 |
ai (Vercel AI SDK) | ^6.1.0 | v7.0.154 | 240 | HIGH | AI SDK v7 replaces maxSteps with stopWhen(stepCountIs(n)) and parameters with inputSchema | npm install --save-exact ai@7.0.154 @ai-sdk/openai@4.0.103 |
openai | ^6.12.0 | v7.45.0 | 43 | MEDIUM | v7.x tightens strict: true tool schemas and standardizes client.responses.create | npm install --save-exact openai@7.45.0 |
@playwright/mcp | npx -y @playwright/mcp@latest | v0.0.83 | 14 | HIGH | Unpinned @latest risks cold-boot browser_snapshot ref changes | Pin @playwright/mcp@0.0.83 in .cursor/mcp.json |
Recommended .cursor/mcp.json Update
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@playwright/mcp@0.0.83", "--headless"]
},
"postgres": {
"command": "npx",
"args": ["-y", "@bytebase/dbhub", "--transport", "stdio", "--dsn", "${DATABASE_URL}"]
},
"agentbreakwatch": {
"url": "https://agentbreakwatch.com/api/mcp"
}
}
}
Want me to apply these exact version pins to .cursor/mcp.json and package.json?

Connects securely to your tools. The creator never sees your data.
About this skill
Stop Silent MCP Schema Drift Before It Breaks Production
AgentBreakWatch Stack and MCP Drift Auditor teaches your coding agent how to discover and audit every dependency manifest (package.json, requirements.txt, pyproject.toml) and Model Context Protocol config (.mcp.json, .cursor/mcp.json, claude_desktop_config.json) in your workspace against the live AgentBreakWatch index of 258 AI packages (32 core AI Agent SDKs and 226 MCP servers on npm and PyPI).
What It Does
- Discovers Your Full AI Stack: Locates JavaScript/TypeScript manifests, Python dependency files, and IDE/agent
.mcp.jsonfiles while automatically redacting any environment credentials before analysis. - Queries Live Registry Telemetry: Connects to the free AgentBreakWatch MCP server (
https://agentbreakwatch.com/api/mcpornpx -y github:joeyg6393/agentbreakwatch-mcp) or the read-only/api/checkendpoint to compare your installed versions against live npm and PyPI versions, 90-day release velocity, and stability scores (0 to 100). - Flags Archived
@modelcontextprotocol/server-*Packages: Detects frozen and archived early reference servers (server-github,server-postgres,server-puppeteer,server-brave-search,server-redis) and provides drop-in configurations for their official vendor-maintained replacements. - Eliminates Cold-Boot
@latestFailures: Identifies every unpinnednpx -y @org/pkg@latestand floatinguvxinvocation in.mcp.jsonand generates exact version-pinned replacement blocks. - Highlights Breaking SDK Boundaries: Alerts you when your stack crosses high-risk major boundaries across
@modelcontextprotocol/sdkv2,ai(Vercel AI SDK v7),openaiv7,@langchain/langgraphv1.5,pydantic-aiv2,fastmcpv4, and@anthropic-ai/claude-agent-sdk.
How It Works
- Reads local workspace manifests in read-only mode.
- Calls
check_stack_drift,get_package_stability,list_volatile_packages, andget_archived_mcp_replacementson the AgentBreakWatch MCP server. - Outputs a severity-ranked Markdown remediation table, ready-to-apply
.mcp.jsonsnippets, and exactnpm install --save-exact/uvxpin commands.
Where It Stops
- Never reads
.envfiles or transmits raw API tokens. - Does not overwrite
package.jsonor.mcp.jsonwithout your confirmation.
Requires external API/MCP connectivity for live data. Does not process private credentials or .env files. Focused only on the 258 tracked AI/MCP packages in the AgentBreakWatch index.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 20 days ago
- Free with an Agensi account
Requires