MDMA

MDMA (Markdown Document with Mounted Applications) extends Markdown with interactive components declared in fenced ```mdma blocks. When an LLM knows the MDMA spec, it responds with structured, validated, renderable UI — forms, tables, approval gates, charts, callouts, webhooks — instead of free-form text. A regular Markdown file becomes an interactive application.

Why it exists: AI conversations today are plain text. Every product that wants an LLM to do something — collect input, confirm an action, show data — has to parse free-form output and hand-build UI for it. MDMA replaces that with one predictable, schema-validated format the model already knows how to emit.

What's in the box:

• 9 component types — form, button, tasklist, table, chart, callout, approval-gate, webhook, thinking
• Deterministic parsing — Markdown + YAML, no runtime JS in documents
• remark-based parser — plugs into the unified/remark ecosystem
• React renderer — @mobile-reality/mdma-renderer-react, all 9 types with hooks
• React Native renderer — the same components rendered as native iOS/Android UI (v1 in progress)
• Validator — 17 lint rules + auto-fix strategies that repair invalid LLM output
• Prompt-pack — model-specialized system prompts (OpenAI, Anthropic, Google, xAI)