An implementation path for both sides
The official Poppy website publishes documentation. It does not run the protocol APIs described by the specification. Companies and personal agents implement their respective sides; example domains are placeholders.
Company side
Choose the execution paths you support: APIs, web pages or a company agent. Map existing account permissions to a dedicated OAuth issuer and make discovery metadata and issuer metadata validate each other.
Implement session creation, renewal and token validation. Offer the sign-in methods and scopes you actually support. Connect business operations to existing authorization: recognizing an agent and allowing it to act for a specific user are separate decisions.
Personal agent side
Publish HTTPS client metadata and public keys, fetch company discovery safely, and verify identity. Assign each user a separate, stable, opaque identifier per company. Do not use an email address or a hash of personal information.
Start a signed-out Session using the company’s advertised capabilities. Request user sign-in when account access is needed, store the resulting Account Token securely, and obtain appropriate Session Tokens for later work.
A first end-to-end check
| Stage | Evidence of success |
|---|---|
| Discovery | Domain and issuer checks pass in both directions |
| Session | The company authenticates the agent and issues applicable credentials |
| Sign-in | User authorization matches granted scopes |
| Execution | A read-only operation returns the expected result |
| Revocation | Revoked account approval cannot obtain further account access |
Start with a read-only task before adding writes, browser sessions and conversations. Use the official draft for exact requests, algorithms and errors; a conceptual flow is not a wire format.