live · v0.2.1 · MCP · MIT
agent-build-sanity
A small MCP server for coding agents. Catches dead model IDs, dependencies with open CVEs, and end-of-life runtimes — the things training data is months stale on — before the agent ships broken code.
Connect
Add to your MCP client config and restart.
stdio-only clients: npx mcp-remote https://dev.miralinhart.com. Raw JSON inspection: curl -X POST https://dev.miralinhart.com -d '…' (full request shape in the source).
Three tools
check_model_currency
saves ~3,000 tokensPass a provider and model ID. Returns deprecation status, days until shutdown, and the named replacement. Sources deprecations.info.
check_dep_health
saves ~5,000 tokensPass an ecosystem and package — npm, pypi, cargo, go, maven, nuget. Returns latest stable, target freshness, deduplicated CVE/GHSA advisories ranked by severity, risk flags. Stitches deps.dev and OSV.dev.
check_runtime_eol
saves ~2,000 tokensPass a runtime, framework, OS, or database. Returns active-support status, EOL date, recommended LTS target. Fully retired products get curated successor names. Sources endoflife.date.
Three output modes
All tools accept mode={concise | json | pretty}. Default concise returns verdict + key facts only — saves ~80% response tokens on the common path. json returns the full structured data block. pretty returns markdown formatted for harness rendering — Claude Desktop, Cursor, Codex lay it out as a card.
Why
Every model has a training cutoff. The world keeps moving past it. By the time the model has been trained, evaluated, deployed, and you're using it, the cutoff is at least three months stale.
Inside that window: provider deprecations land, CVEs get published, runtimes hit EOL, packages get hijacked, defaults change. A single MCP call that's current about any of this saves the agent from confidently picking an answer the world abandoned six months ago.
Free. Stateless. No caller input retained. Cloudflare Worker hosted. Five-minute integration. Source on GitHub, MIT.