BlueVetaUpright is an API that detects whether a product photo is sideways or upside-down and tells you how to fix it. We just published it on the official Model Context Protocol registry (com.blueveta/upright), so any MCP-compatible AI agent — Claude, Cursor, etc. — can call it directly.
One design question came up before we built this: doesn't an MCP server just give AI agents a free way around the paid API? We went with BYOK (bring your own key) specifically to avoid that — the MCP tool calls the exact same paid /detect endpoint a normal REST integration would use, with the caller's own BlueVetaUpright API key. There's no shared key and no route into the free, IP-capped demo. An agent using this tool pays exactly the way a developer calling our REST API directly would.
On accuracy: our published benchmark (methodology and full breakdown at the link below) is 99.05% on clean, upright-or-90-degree-rotated photos, 96.52% on photos with any amount of added tilt, and 83.63% on the hardest cases we test (tilted up to ±40 degrees with no clean quadrant to anchor on). We also ran a real head-to-head against general-purpose vision models on this specific task and came out ahead.
If you're building an agent that handles product photos — e-commerce listings, marketplace sync, bulk photo pipelines — this means it can catch and fix orientation problems without you writing custom integration code.
Free demo, no signup: https://blueveta.com/upright
Research/benchmark: https://blueveta.com/upright/research
MCP registry entry: com.blueveta/upright
Top comments (0)