What is MDAN?

One page for humans and agents.

MDAN (Markdown Action Notation) is a shared notation for interactive pages that stay readable and actionable for both humans and agents, from the same source, across any interface.

Same page. Same actions. Same experience.

What Makes It Different

In a typical AI product stack, page content, tool definitions, follow-up actions, and browser UI often live in separate layers.

MDAN pulls those pieces back into one readable and actionable page-shaped application surface:

Markdown carries the content people and agents read
executable operations are declared alongside that content
server responses return Markdown fragments that can drive the next step
the same app can negotiate Markdown for agents and HTML for browsers

Who It Is For

MDAN is a strong fit for teams building:

agent-facing apps
internal tools that need both browser and agent access
workflow or skills apps with multi-step page interaction
docs-like or form-like products that should stay readable in source

When To Use MDAN

Use MDAN when:

one app needs to work well for both humans and AI agents
you want a readable source format instead of splitting everything across templates, prompts, and JSON
your interaction model is page-shaped and evolves step by step
you want one server app to serve both Markdown and HTML from the same logic

When Not To Use MDAN

MDAN is probably not the right fit when:

your product is only a conventional browser app
your backend is primarily a JSON API with no page-shaped interaction
your UI depends on a large client-only state model
you do not need an agent-readable interaction surface

Node and Bun Support

MDAN officially supports Node and Bun.

use Node if you want the most established host baseline
use Bun if you want a Bun-native starter and toolchain entry
keep the same app model in both cases