Finding & Reaching Entities for Digital Agents
The open protocol that unifies 12+ agent discovery protocols (MCP, A2A, OpenAPI, llms.txt, Schema.org, and more) into a single JSON file. Add it to your domain — every AI agent can instantly discover and interact with you.
Website · Whitepaper · Auto-Install · Validator
AI agents are becoming the primary interface between humans and the internet. But today, when an agent needs to interact with a business, API, or service, it faces:
- No standard discovery. How does an agent find out what a website can do?
- Fragmented protocols. MCP, A2A, ADP, llms.txt — each solves a piece, none solves the whole.
- No trust signals. How does an agent know if an entity is legitimate?
- Opaque pricing. How does an agent compare services or execute transactions?
- No rules of engagement. What is an agent allowed to do? What data can it store?
One file. Full story.
Add a fred.json to your domain, and any AI agent can instantly understand:
| Layer | Question | Example |
|---|---|---|
| Identity | Who are you? | Name, type, contacts, jurisdiction |
| Capability | What can you do? | Create CVs, search events, process payments |
| Interaction | How do I talk to you? | REST API, MCP server, GraphQL endpoint |
| Trust | Why should I trust you? | Business verification, uptime, compliance |
| Pricing | What does it cost? | Free tier, $12/mo Pro, enterprise |
| Context | What should I know? | FAQs, limitations, SLA, use cases |
| Discovery | Who else is related? | Partners, integrations, alternatives |
| Policy | What are the rules? | Rate limits, data handling, attribution |
FRED doesn't replace MCP, A2A, or llms.txt — it binds them together into one implementation surface.
{
"fred": "0.1",
"identity": {
"name": "Your Company",
"type": "organization",
"description": "What you do in one sentence",
"url": "https://yourcompany.com"
}
}That's a valid FRED file. Only fred and identity are required. Everything else is optional and progressive.
Place the file at:
https://yourcompany.com/.well-known/fred.json
Or at the root:
https://yourcompany.com/fred.json
node validate.js your-fred.jsonFRED Protocol Validator v0.1
──────────────────────────────────────────────────
Entity: Your Company (organization)
Version: 0.1
Layers: identity
VALID — fred.json passes validation
Conformance: Level 1 — Basic
──────────────────────────────────────────────────
Add more layers as you need them:
{
"fred": "0.1",
"identity": { ... },
"capabilities": [
{
"id": "search-products",
"name": "Search Products",
"description": "Full-text search across product catalog",
"category": "search",
"inputs": [
{ "name": "query", "type": "string", "required": true, "description": "Search query" }
],
"outputs": [
{ "name": "results", "type": "array", "description": "Matching products" }
]
}
],
"interaction": {
"protocols": [
{ "type": "rest", "url": "https://api.yourcompany.com/v1" },
{ "type": "mcp", "url": "https://mcp.yourcompany.com" }
]
}
}| Level | Name | Requirements |
|---|---|---|
| 1 | Basic | Valid fred.json with identity layer |
| 2 | Interactive | Level 1 + capabilities + at least one interaction protocol |
| 3 | Complete | Level 2 + trust verification + agent policy |
Two complete, Level 3 examples are included:
A SaaS CV builder — demonstrates product capabilities, OAuth2 + MCP interaction, pricing tiers, trust signals, and comprehensive agent policy.
A developer community — demonstrates community entity type, event discovery, member search, resource curation, and community-driven trust.
Agent receives user request: "Find me a CV builder in Singapore"
|
v
Agent discovers fred.json at freddys.io/.well-known/fred.json
|
v
Agent reads identity → "AI-powered CV builder, Singapore, 15K users"
|
v
Agent reads capabilities → "create-cv, optimize-cv, generate-cover-letter"
|
v
Agent reads interaction → REST API + MCP server available
|
v
Agent reads trust → Verified business, 4.7/5 rating, 99.95% uptime
|
v
Agent reads pricing → Free tier available, Pro at $12/mo
|
v
Agent reads policy → Can create CVs, must attribute, session-only data
|
v
Agent connects via MCP and creates a CV for the user
FRED is the binding layer. It doesn't compete — it connects.
┌─────────────────────────────────────────────┐
│ fred.json │
│ The single entry point for AI agents │
├─────────────────────────────────────────────┤
│ │
│ ┌─────────┐ ┌─────────┐ ┌──────────────┐ │
│ │ MCP │ │ A2A │ │ OpenAPI │ │
│ │ Server │ │Endpoint │ │ Spec │ │
│ └─────────┘ └─────────┘ └──────────────┘ │
│ │
│ ┌─────────┐ ┌─────────┐ ┌──────────────┐ │
│ │llms.txt │ │ REST │ │ WebSocket │ │
│ │ │ │ API │ │ Endpoint │ │
│ └─────────┘ └─────────┘ └──────────────┘ │
│ │
└─────────────────────────────────────────────┘
The complete protocol specification is in fred-spec.md. It covers:
- All 8 layers in detail with schemas and examples
- File location and discovery rules
- Security considerations
- Versioning
- Conformance levels
- IANA considerations
The machine-readable schema is at fred-schema.json. Use it with any JSON Schema validator:
const Ajv = require("ajv/dist/2020");
const schema = require("./fred-schema.json");
const ajv = new Ajv();
const validate = ajv.compile(schema);
const valid = validate(yourFredJson);A zero-dependency reference validator is included:
# Validate a file
node validate.js path/to/fred.json
# JSON output (for CI/CD)
node validate.js path/to/fred.json --json
# Read from stdin
cat fred.json | node validate.js --stdinFREDProtocol/
├── fred-spec.md # Full protocol specification
├── fred-schema.json # JSON Schema (Draft 2020-12)
├── validate.js # Zero-dependency reference validator
├── web/ # hungryhuman.co/fredprotocol — Next.js website
│ ├── app/ # Pages: home, whitepaper, auto-install, validator
│ └── components/ # Shared UI components
├── claude-skill/ # Claude Code plugin (auto-detects deploys)
│ ├── plugin.json # Plugin manifest with PostToolUse hook
│ ├── hooks/ # Auto-detection hook
│ └── skills/ # FRED generation skill
├── examples/
│ ├── freddys.io/fred.json # SaaS product example (Level 3)
│ └── claudecode.sg/fred.json # Developer community example (Level 3)
├── IMPLEMENTING.md # Platform-specific setup guides
├── GOVERNANCE.md # How the spec evolves
└── LICENSE # Apache 2.0
The FRED Protocol is open source and community-driven. See GOVERNANCE.md for how the spec evolves and how to contribute.
Ways to contribute:
- Implement fred.json for your domain and share it
- Build tools (validators, generators, crawlers)
- Propose spec changes via GitHub Issues
- Write integrations for AI agent frameworks
- Translate the spec
Is this an Anthropic project? No. FRED is an independent open-source protocol created by Freddy Lim.
Why not just use Schema.org? Schema.org is optimized for search engine crawlers. FRED is agent-first — designed for AI agents that need to take action, not just index content.
Do I need to implement all 8 layers?
No. Only identity is required. Start there and add layers as you need them.
What about authentication?
FRED doesn't handle auth directly. It points agents to your auth endpoints via interaction.protocols[].auth. Use whatever auth system you already have.
How does trust work? Trust in FRED is not a single score. It's a collection of independently verifiable claims — business registrations, uptime records, compliance certifications, user ratings. Agents can weigh these signals according to their own trust models.
Will you register this with IANA?
The .well-known/fred.json URI and application/fred+json media type will be submitted for registration as the spec matures beyond draft status.
Apache 2.0 — see LICENSE.
Created by Freddy Lim
FRED Protocol is not a product. It's infrastructure for the agentic web.