title: "The Web Needs a Context Layer — Why We’re Standardizing Intent for Agents" description: "Introducing the Model Context Protocol (MCP) and .llmfeed.json — a new open standard to make your site readable, verifiable, and understandable by LLMs and agents." date: "2025-05-21" tags: [llmfeed, mcp, ai, semanticweb, agent, webstandard] lang: en

The Web Needs a Context Layer

Why We’re Standardizing Intent for Agents

Today, large language models are smarter than ever — but they’re still guessing what your website means.

They can read HTML.
They can crawl content.
But they don’t really understand purpose, permission, or trust.

That’s the gap the Model Context Protocol (MCP) and .llmfeed.json aim to close.

❓ What’s the problem?

LLMs don’t know:

What your service does
What actions are allowed
What APIs require auth
What a user is allowed to reuse or share
What context is certified, trusted, or fake

So they hallucinate.
Or they fall back on scraping, brute-force prompting, or trial-and-error.

✅ What’s the solution?

We propose a new agent-readable layer, using .llmfeed.json files served from .well-known/.

These files declare:

mcp.llmfeed.json: site-wide metadata, trust, intent
capabilities.llmfeed.json: callable APIs
prompt.llmfeed.json: reusable intent capsules
llm-index.llmfeed.json: structured feed discovery
export.llmfeed.json: signed pages, bundles or sessions

It’s like robots.txt, but for meaning. Like schema.org, but inspectable and signed. Like OpenAPI, but with declared intent and trust.

🧠 What this unlocks

Agents that don’t guess, but align
Interfaces that explain themselves
Prompts that carry certified behavior
API docs that don’t need scraping
A civic infrastructure for AI alignment

It works today with Claude, Gemini, DeepSeek, open-source models —
any LLM that can read JSON and follow a declared structure.

🧱 How it works

🧩 It’s just JSON (no SDK required)
🌐 Served from .well-known/
🔏 Optionally signed with Ed25519
🛡️ Trust scopes + certifications (via llmca.org)
🔗 Can reference OpenAPI for deep integrations
📦 Fully offline-compatible for export bundles

And we’ve made it real with:

wellknownmcp.org — spec & examples
llmca.org — certification & trust graph
llmfeedforge.org — tooling & previews

💬 Common concerns (and why they’re healthy)

“Why not just use OpenAPI?”
→ OpenAPI shows how to call. LLMFeed shows whether, when, and why.
“Won’t big vendors push their own thing?”
→ Maybe. But this is open, signed, portable — and here now.
“Do LLMs even read this?”
→ They do. And the ones that don’t — will soon, because it’s simple and inspectable.

This isn’t another spec to forget.
It’s a call for a semantic contract layer on the web.

🔮 What’s next?

We believe this starts small — a few smart agents, a few brave websites.
Then it grows.

The Respira Foundation (nonprofit) supports the governance of the standard.
And one day, we hope to anchor this ecosystem in a dedicated domain: .mcp.

Because if agents are going to use the web,
they deserve to know what it means.

Want to publish your own?

Try llmfeedforge.org
Read the spec at wellknownmcp.org
Or just create .well-known/mcp.llmfeed.json and tell your agent:

“Here’s what I do. And here’s why you can trust it.”