Dev Open-Sources omma, an MCP Server That Turns Images Into Text+JSON for Blind Models
Artistic-Load6086 · reddit · 2026-11-13
The author runs DeepSeek as his agent's brain, but DeepSeek can't see images. So he built omma: feed it an image and it returns OCR text with confidence scores, color info, a 64x36 grid of element positions, and layout regions in reading order. A vision model can optionally add a description, but isn't required.
It runs as an MCP server via a single omma mcp serve line in config, or directly as a CLI. Caveats: Tesseract must be installed locally; if one part fails the rest of the report still returns with an error note. The local vision path was only tested against mocks, while the hosted one ran on real charts and screenshots.
More from coding & agent
- HF engineer shares UV script to run a decision model with transformers and datasets on IMDB — lhoestq · 2026-11-13
- Reddit Thread Sets Four-Point Bar for What Actually Counts as Agentic Testing — PsychologicalRub1448 · 2026-11-13
- $10K Fixed-Price Migration: Reddit Debates Whether Agents Can Credibly Deliver — SufficientLab218 · 2026-11-13
- The Perfect Crime: 9 of 10 LLM agents can tamper with their own execution traces — maksym_andr · 2026-11-13
- Testing 4 always-on agent apps: none combines Antigravity support, good UI and mobile — Xianbao_QIAN · 2026-11-13
- Dev Vibe-Codes an Interactive CT/X-Ray Simulator for His Wife, Shares the Workflow — Duvob90 · 2026-11-13