Optical Context MCP

This server turns long, visually dense PDFs into compressed image sets that preserve layout and structure. It wraps Mistral OCR behind three MCP tools: compress_pdf reads a local file and runs extraction plus recomposition, get_job_manifest returns metadata about the packed output, and get_packed_images fetches the resulting PNGs in batches. It bundles an adaptive sizing model that attempts to estimate figure dimensions intelligently, falling back to fixed sizing when the ML runtime is absent. Reach for it when you're working with operating manuals, scanned handbooks, or technical documents where visual grouping matters more than clean text extraction. Artifacts live in the system temp directory and jobs are identified by UUID for follow-up retrieval.