I built a free VIN decoder on public gov data, then made it callable by AI agents (MCP). Here's the build + what I learned.

Quick background: paid VIN report services charge $25–45, but a lot of the core data — vehicle specs and open safety recalls — is free US government data (NHTSA APIs, no key, no cost). That gap bugged me, so I built CheckMyVIN: paste a VIN, get specs + open recalls + a plain-English AI summary, free, no signup.

Three things I learned building and distributing it solo:

1. Honesty is a feature, not a constraint. When you don't have real data, say so. My tool returns "tire size varies by trim — check the door jamb sticker" instead of guessing. Sounds minor, but it's the difference between a tool people trust and one they catch lying once and never return to. EVs get battery/warranty info, not "what oil does your Tesla need" — most VIN tools get this embarrassingly wrong.

2. "Callable by an agent" is a different game from "described by AI." I added an MCP server, so you can ask Claude (or any MCP-capable client) "check recalls on this VIN" and it calls the tool directly, mid-conversation. Building that forces honesty even harder — an agent surfaces garbage output straight to the user, no marketing layer in between.

3. Distribution as a solo founder is layered, not one big launch. This week: official MCP Registry, a dev.to writeup, Product Hunt, directories. Each is small; together they compound. The MCP angle specifically has almost zero competition right now.

It's free and ad-free (I'm testing affiliate links for the paid-data stuff I don't offer, like accident history — honest about what it is and isn't).

If you want to see it: https://checkmyvin.net — and the MCP server is at checkmyvin.net/api/mcp.

Happy to answer anything about the build, the NHTSA data, or the MCP side. What would you want a tool like this to do that it doesn't yet?