MDMA (Markdown Document with Mounted Applications) is an MIT-licensed TypeScript library that extends Markdown with interactive components, so an LLM can answer with forms, tables, and approval gates that a host application renders inline instead of leaving the user to act on plain text.

What it is

MDMA extends Markdown with interactive components defined in fenced mdma code blocks, so a regular Markdown file becomes an interactive application. It is written in TypeScript under the MIT licence, and its topics place it in the LLM and agent tooling ecosystem: llm, ai-agents, agent-skills, claude-skill, mcp, mcp-server, model-context-protocol, generative-ui, remark-plugin, react, and interactive-markdown. The GitHub description calls it "interactive documents from Markdown" that extend Markdown with forms, approvals, and webhooks, built for next generation apps.

The problem it replaces is specific: AI conversations today are plain text, and the user reads a response and then manually acts on it. MDMA closes that gap by letting the model answer with structured components rather than prose. The README states the trade in terms of what it removes — no custom UI per use case, and no parsing free-form text, because the AI generates structured, validated components the frontend renders instantly. The worked example is a patient intake document: an mdma block with type: form and id: intake-form declares fields such as patient-name (type text, required: true, sensitive: true) with onSubmit: submit-intake, followed by a type: button block with id: submit-btn, variant: primary, and onAction: submit.

Key capabilities

• Fenced mdma code blocks that extend ordinary Markdown, each declaring a type such as form or button and a stable id.
• Form fields carrying metadata the renderer and backend can act on: name, type (text, email, textarea), label, required, and sensitive.
• Action wiring through onSubmit and onAction, so a component dispatches to a named target instead of relying on free-form text.
• Component classes named in the description and tagline: forms, tables, approval gates, and webhooks.
• Integration surfaces for the Model Context Protocol and adjacent tooling, reflected in the mcp, mcp-server, model-context-protocol, remark-plugin, and react topics.
• An authoring prompt layer: MDMA_AUTHOR prompt variants live under packages/prompt-pack/src/prompts/mdma-author/, and a hosted MDMA-IL model is published on Hugging Face as MobileReality/mdma-gemma4-26b-dsl-unsloth-v1.

Who uses it and how

• Frontend developers building React applications that render model output, using the remark-plugin path to turn mdma blocks into live components.
• Teams running MCP servers, exposing MDMA authoring or rendering over the Model Context Protocol so an agent emits valid components.
• Agent and Claude skill users who want a conversation to end in a completed form or an approved step rather than a paragraph of instructions.
• Domain workflows with structured intake and sensitive data, as shown by the README's patient intake form with sensitive: true fields.
• Prompt contributors, who add a variant under packages/prompt-pack/src/prompts/mdma-author// and open a pull request; maintainers run the eval suite and add the result to the MDMA_AUTHOR matrix.

Getting started

The README excerpt provided shows no install or run command, no package name, and no Docker image; the entry points it does expose are the Website, Live Demo, and Docs links. Contributors are directed to add a prompt variant under packages/prompt-pack/src/prompts/mdma-author// and open a pull request.

How it compares

The facts list no commercial products that MDMA replaces, so no licence, self-hosting, or cost comparison against paid alternatives can be made. Its nearest named neighbours are the Model Context Protocol ecosystem and Markdown/remark plugin tooling, which appear here as integration surfaces rather than competitors; on the evidence given, it stands alone in this registry.

When to use it — and when not

The facts mention no infrastructure a self-hoster must operate, so the burden is whatever the host application already runs plus the LLM path. Teams that need a widely adopted dependency with a documented published package should look elsewhere for now: the README excerpt shows no install command, the project has 67 stars and 2 forks, several prompt matrix entries remain TBD, and gpt-5.4 carries a non-deterministic output duplication bug in multi-turn, custom-prompt, and flow evals in roughly 7–15% of runs.

# Patient Intake

```mdma
type: form
id: intake-form
fields:
  - name: patient-name
    type: text
    label: "Full Name"
    required: true
    sensitive: true
  - name: email
    type: email
    label: "Email"
    required: true
    sensitive: true
  - name: reason
    type: textarea
    label: "Reason for Visit"
    required: true
onSubmit: submit-intake
```



```mdma
type: button
id: submit-btn
text: "Submit Intake Form"
variant: primary
onAction: submit
```