Overview:
MDMA is an open-source specification and library that extends Markdown with interactive, structured components. It enables AI-generated user interfaces—such as forms, tables, approval gates, and webhooks—to be rendered inline within a Markdown document. Instead of parsing free-form text, applications receive predictable, validated components from an LLM. The project is designed for developers building AI-powered apps or chat interfaces that need to move beyond plain-text responses into actionable, rendered interactions. It provides tooling for prompt building, document validation, and system integration.
Core Features:
Structured Interactive Components: Extends Markdown with nine component types (form, button, tasklist, table, chart, callout, approval-gate, webhook, and thinking block) defined in fenced code blocks by a YAML schema.
Deterministic Parsing: Parses Markdown and YAML without executing runtime JavaScript, ensuring predictable output for applications.
PII Protection: Provides automatic detection and redaction of personally identifiable information via hashing, masking, or omission.
Audit Trail: Maintains an append-only event log with tamper-evident hash chaining for document interactions.
Policy Engine: Supports allow/deny rules configured per action and environment for controlling component behavior.
Validation & CLI Tooling: Includes a CLI for validating MDMA documents, with options for auto-fixing issues and providing JSON output, plus a prompt builder for creating AI system prompts.
Use Cases:
Developers building LLM-based chat interfaces: Integrate structured, interactive responses (such as intake forms or approval steps) directly into AI-powered conversational apps, replacing plain-text replies.
AI prompt engineers: Use the prompt builder and system prompt generation to create structured prompts that enable LLMs to output MDMA components reliably.
Automation and workflow developers: Embed interactive webhook triggers and approval gates in generated documents to create actionable, step-based processes.
Why It Matters:
MDMA offers a structured, predictable layer between an AI model and the user interface without requiring custom UI per use case. Its deterministic parsing and YAML-based component schema allow frontends to render components instantly without free-form text parsing. The validator and policy engine provide tooling for ensuring document integrity and controlling actions in various environments. It is actively used in production by its maintainer, Mobile Reality, primarily in fintech and proptech projects.