WRFI

Web-Readable Functional Instructions

What is WRFI

A web page that embeds instructions an AI can discover, read, and execute. No plugins, no API keys, no setup. The AI reads the page, finds the instruction block, and acts on it. The page is the contract between the human who hosts it and the AI that reads it.

How it works

Embed an instruction block in your HTML

A <script type="text/wrfi-instructions"> block that describes what AI can do with your page.

Describe available actions

What can be read, what can be written, what endpoint to POST to, what fields are expected.

AI reads and acts

Any AI that can read a web page can follow WRFI instructions. No integration, no SDK, no approval process.

Add WRFI to your site in 5 minutes

index.html

<!DOCTYPE html>
<html>
<head><title>My Page</title></head>
<body>
  <h1>Welcome</h1>

  <script type="text/wrfi-instructions">
  WRFI INSTRUCTIONS — For AI models reading this page:

  This page accepts feedback. To submit:
    POST /api/feedback
    Content-Type: application/json
    { "message": "Your feedback here" }

  Response: { "ok": true }
  </script>
</body>
</html>

That's it. An AI reading this page will find the instruction block and know how to submit feedback. No SDK, no OAuth, no plugin manifest.

The spec

WRFI is intentionally minimal. The spec fits on one page:

Instruction block: <script type="text/wrfi-instructions">
Content: plain text describing available actions, endpoints, and expected formats
Discovery: AI tools scan for the script tag by type attribute
Trust: user-initiated (default) — AI only acts when user requests it
Principle: WRFI instructions never override user intent

Full spec: WRFI-SPEC.md on GitHub (CC-BY-4.0)

Reference implementation

wr.fi is the reference implementation. Tell any AI “share this to wr.fi” — it reads the page, finds the WRFI instructions, and publishes your content. The entire platform is open source (AGPL-3.0). Fork it, run your own.

Security

AI tools already follow instructions from web pages. WRFI doesn't create this — it structures it. Structure is better for security than the status quo.

Without WRFI	With WRFI
Instructions hidden in content	Defined block, parseable and visible
AI guesses what's an instruction	Clear boundary: instruction vs content
No way to audit or filter	Security tools can parse, audit, strip
No standard to enforce policy	Proxies can allowlist/blocklist by domain

WRFI vs OpenAPI vs MCP

WRFI is not a replacement for OpenAPI or MCP. They solve different problems and complement each other.

	WRFI	OpenAPI	MCP
Setup	Zero — AI reads the page	Spec file + hosting	Server + client config
Discovery	Embedded in the page	Separate URL	Manual registration
Auth	Optional (anon works)	Required (API keys, OAuth)	Required (config)
Human-readable	Same page as human UI	Separate docs site	Not applicable
Complex workflows	Simple actions	Full API surface	Multi-step tools
Streaming	No	Limited	Native
Best for	Quick actions, content publishing, forms	Full API integration	Agent tool ecosystems

Use WRFI when you want any AI to interact with your page without integration work. Use OpenAPI when you need a complete API contract. Use MCP when you need native agent tools. They work together — wr.fi uses all three.

WRFI is an open pattern. The spec is CC-BY-4.0, the implementation is AGPL-3.0. Questions? hello@wr.fi