An auth protocol for AI agents, distributed as markdown

Encrata publishes an auth.md file at the root of its domain. Your agent reads it to learn how to authenticate with an API key, then calls the API directly - yours to read, parse, and act on.

Declare an intent once. Your agent reads the auth file, authenticates with the user's API key, and runs email lookups from the same source.

> curl https://encrata.com/auth.md

> # agent reads, authenticates, calls


200 OK - agent authenticated via auth.md


The file

The file is the contract. Every detail an agent needs - authentication flow, API key format, available endpoints, credit system, MCP config, and error codes - is a section of one markdown file. Fetch what you need; the agent never asks the user.

Each section is structured for machine parsing. The agent reads auth.md once at startup. Local caching is fine - the file changes only on API updates.

> GET /auth.md


# auth.md

You are an agent. Ask the user for their Encrata

API key. Never ask for their password.


## Getting a key

app.encrata.com/settings/api-keys


## Using the key

Authorization: Bearer enc_live_xxxxxxxxxxxx


## Available now

POST /api/email/lookup - Full profile

POST /api/email/validity - Deliverability

POST /api/email/smtp-verify - Mailbox check

POST /api/email/breaches - Breach exposure

POST /api/email/password-breaches - Password leak

full file at encrata.com/auth.md - MCP config and error codes included. Phone, IP, domain and company lookups are coming soon and not callable yet.


MCP as a peer of HTTP

Agents that speak MCP can skip HTTP entirely. Add the Encrata server to your MCP config and every lookup becomes a tool call. The auth.md file documents both paths - HTTP and MCP - so your agent picks whichever it supports.

> cat mcp.json


{

"mcpServers": {

"encrata": {

"url": "https://mcp.encrata.com",

"headers": {

"Authorization": "Bearer enc_live_xxxx..."

}

}

}

}

Tools: lookup, email_validity, smtp_verify, email_breaches, password_breaches


Works with

Any agent that can fetch a URL and parse markdown can use auth.md. No SDK required, no special integration - just HTTP GET and string parsing.

Claude Code • Cursor • Windsurf • ChatGPT • CrewAI • LangChain • LlamaIndex • Custom agents


Point your agent at encrata.com/auth.md and it handles the rest. No integration guide, no webhook setup, no client library.