The open-source MDMA protocol extends standard Markdown with typed UI components, allowing large language models to generate structured forms, approval gates, and webhooks that applications can render without custom interfaces or fragile JSON parsing.
WARSAW, Poland — 18.03.2026 — Mobile Reality sp. z o.o. has released MDMA, an open-source Markdown extension designed to transform plain-text LLM outputs into interactive, actionable interfaces. The protocol introduces schema-validated components—including forms, buttons, approval gates, and webhooks—inside standard Markdown streams, solving the persistent last-mile problem that forces users to interpret prose instructions rather than act directly within chat environments. Available on GitHub at https://github.com/MobileReality/mdma and published under the MIT License, MDMA gives developers a deterministic, model-agnostic alternative to brittle JSON or raw HTML payloads.
Large language models have excelled at generating prose, yet they leave users with instructions to act upon rather than actions to complete. Attempts to bridge this gap with raw HTML or JSON payloads have proven fragile; cheaper and faster models such as Gemini 2.5 Flash routinely corrupt nested structures, producing unparseable output that breaks the frontend. The result is a forced dependency on premium LLMs and a poor user experience. MDMA addresses this by extending Markdown with nine typed, schema-validated elements embedded inside fenced mdma code blocks. Because Markdown is already native to chat and documentation platforms, models handle the syntax reliably, and developers avoid building a custom interface for every new conversational workflow.
The MDMA ecosystem is published as ten modular packages on npm under the Mobile Reality organization. At its foundation sits @mobile-reality/mdma-spec, which defines strict schemas for components such as Form, Button, Tasklist, Table, Chart, Callout, Approval Gate, Webhook, and Thinking blocks. The @mobile-reality/mdma-parser processes these blocks as a remark markdown plugin, while @mobile-reality/mdma-runtime manages reactive state and includes built-in PII redaction to protect sensitive data before it leaves the client environment. A companion React renderer allows frontend teams to mount interactive surfaces with minimal integration effort. Together, the parser, static validator, and deterministic fixer achieve 98% deterministic UI reliability across leading model families including GPT, Claude, Gemini, and Llama, enabling organizations to deploy smaller, faster, and less expensive models in production.
MDMA's architecture directly reduces operational costs by removing the need for high-token premium models to generate stable user interfaces. Its deterministic parsing also means fewer retries and reduced latency at runtime. Mobile Reality uses MDMA as the engine behind its proprietary AI Headless Forms commercial solution, which is actively deployed in fintech and proptech environments. The headless design gives enterprises full control over branding and layout while ensuring that collected data remains within the client's own infrastructure. This combination of cost efficiency, design independence, and strict data privacy provides a concrete return on investment for organizations scaling conversational AI.
"Standard JSON-in-chat approaches fail because smaller models frequently hallucinate structure, drop brackets, or break nesting, which shatters the UI at runtime without warning," said Marcin Sadowski, CTO of Mobile Reality. "MDMA anchors every interactive element inside Markdown code blocks that these models already generate millions of times per day. Our parser, static validator, and deterministic fixer work in sequence to deliver 98% reliability across GPT, Claude, Gemini, and Llama. That means teams can ship on fast, inexpensive hardware instead of burning tokens on frontier models just to get a form to render."
"Enterprises want to adopt LLMs without exposing sensitive data to third-party UI layers or inflating their cloud bills," said Matt Sadowski, CEO of Mobile Reality. "MDMA keeps the interface layer deterministic and the data inside the client's stack, which slashes token costs and preserves privacy. Our AI Headless Forms product, built entirely on MDMA, is already collecting data in fintech and proptech workflows. The open-source core invites collaboration, while the commercial platform proves that secure, scalable conversational interfaces can deliver hard ROI."
About Mobile Reality
Mobile Reality sp. z o.o. is an AI automation agency and product software development company based in Warsaw, Poland. Co-founded in 2016 by brothers Marcin and Matt Sadowski, the company builds AI-powered solutions for enterprises, with its commercial platforms deployed in fintech and proptech sectors.
