prompt2mdA Markdown Magic

Turn anything into token-optimized, layout-aware Markdown — and know exactly what it saved you. Fast, easy, reliable, and nothing is ever lost.

Try the studio

Get started

Integrations

Architecture

🧭

Dual-engine routing

Fast path (~0.6s) for HTML/Office/CSV/simple PDFs; high-fidelity TableFormer + OCR for scans and complex tables. Routed by content evidence, never by file extension.

🎯

Token cost as a first-class output

Every conversion returns a report — input/output tokens, compression ratio, per-section counts, and budget verdicts. Numbers you can reproduce, not marketing math.

🗂️

Prompt-cache-aware layout

Stable prefix first, provider-specific breakpoints, volatile content last. Repeat calls cost up to ~90% less on cache-enabled providers.

🔎

Lossless by construction

Originals are stored before compression touches them. Every summarized section carries a p2md:src anchor — retrieve_original returns the byte-exact source.

🤖

Agent-native

MCP server (convert, compress_context, retrieve_original, and an optimize chat-box prompt), a /prompt2md skill, and a batch CLI — all over one shared runtime.

🪶

Degrades gracefully

Zero sidecars configured? Deterministic cleanup and extractive summarization take over, with honest warnings. Nothing hard-fails on textual input.