OPENAPI APPROVED · x-agent-trust is officially registered in the OpenAPI Extensions Registry · Merged 11 April 2026

Official OpenAPI Extension

Build with x-agent-trust. OpenAPI approved now.

The first vendor extension in the OpenAPI registry designed for AI agents. Drop it into any OpenAPI spec. Agents authenticate with signed trust tokens. APIs verify locally. No gateway. No SaaS dependency.

Start Building View in Registry ->

The Extension

3 lines in your OpenAPI spec

By Raza Sharif, CyberSecAI Ltd | MERGED into the OpenAPI Extensions Registry | PR #67 | Discussion #5267

openapi.yaml

# Add to any OpenAPI 3.0 / 3.1 spec
x-agent-auth:
  algorithm: ES256
  trustLevels: [L0, L1, L2, L3, L4]
  issuerKeysUrl: /.well-known/agent-trust-keys

paths:
  /v1/charges:
    post:
      x-agent-trust-required: L2

API server (one line)

const { verifyAgentTrust } = require('mcp-secure');
app.use(verifyAgentTrust({ minTrust: 'L2' }));

Agent side (two lines)

const { AgentIdentity } = require('mcp-secure');
const agent = new AgentIdentity({ trustToken: 'eyJ...' });
await agent.fetch('https://api.stripe.com/v1/charges');

Demo API

4 endpoints, 4 trust levels

Each endpoint requires a minimum trust level. The middleware verifies locally.

Endpoint	Method	Description	Required Trust
`/api/data`	GET	Read data	L1
`/api/charges`	POST	Create payment	L2
`/api/execute`	POST	Execute workflow	L3
`/api/admin`	DELETE	Admin operation	L4

Live Demo

Try it now

Select an agent and an endpoint. Watch the trust verification happen in real time.

Agent Trust Verification

Agent

Endpoint

On the Wire

Real HTTP with the Agent-Signature header

This is the actual wire format: a signed HTTP request with an RFC 8941 structured-field header, verified end-to-end against the JWKS.

Sign + Verify an HTTP Request

Agent

Method + Path

Request body

HTTP REQUEST (signed)

CANONICAL REQUEST (signed over)

VERIFICATION STEPS (legitimate request)

VERIFICATION STEPS (tampered body — amount changed)

How It Works

Local verification. Like JWT.

1. Agent sends request with signed trust token
Agent --> API: POST /api/charges + Agent-Signature header

2. Middleware verifies locally (no remote call)
✓ Verify ECDSA signature (cached public key)
✓ Check token expiry
✓ Read trust_level from payload
✓ Compare against endpoint requirement

3. Allow or deny
Agent trust L2 >= Required L2 --> ALLOW

No gateway. No remote call. No vendor dependency.

Trust Framework

L0 through L4

The API sets the minimum. The token carries the level. The middleware compares.

Level	Meaning	Typical Access
L0	Unknown agent	Rejected
L1	Identity verified	Read-only
L2	Trust established	Read + limited write
L3	Highly trusted	Read + write + execute
L4	Fully trusted	Read + write + execute + admin