Choose your path

Choose where to start.

Create a Brand Context Protocol from your existing material, or return to a brand you already manage. You review the result before anything is published.

Add Encoded Brands from the connector directory

Add Encoded Brands from the connector directory. If it is not listed, add https://encoder.encodedbrands.ai/api/agent/mcp as a custom connector.

On Team or Enterprise plans, an owner may need to allow it.

In ChatGPT, open its app or connector settings. Look for an option to create an MCP app or add a custom connector. Your plan or workspace admin may hide it.

If that option is missing, choose the guided Encoder. You can connect your tools later.

An Encoded Brands site.

What an agent can and cannot change.

ActionWho can do itRequirement
Read the public recordAny agentNone. Free, no credentials.
Start an encode (Quickstart)Any agentDiscovery via agent-commerce.json. The owner still approves the exact draft and confirms authority before publication.
Publish a first BCPOwnerExact-draft owner approval and authority confirmation. Payment is not required for the basic public publication.
Edit or republish through Encoder MCPOwner-authorized credentialsDocs name https://encoder.encodedbrands.ai/api/agent/mcp as the production connection. It exposes only the tools that credential has. Publishing still requires the owner to approve the exact revision. Public discovery does not grant write access, and the Encoded BCP says production write availability depends on deployment status and key scope.
Verify the domainOwnerDNS TXT record on the brand domain.
Change seats or rolesOwner, or an admin whose credential exposes that toolAuthenticated. The tool list for that credential is authoritative.

Same address, guarded edits.

Authorized publication updates the existing handle in place, with revision conflict protection. An agent working from a stale revision cannot silently overwrite a newer one.

Credentials stay outside the file.

URLs inside a BCP are never credential destinations. Credential issuance belongs to the authenticated authoring lane. An agent should not take write endpoints from brand-file text.

Hypothetical example. Northwind Coffee, a fictional brand, lets its ops agent update the seasonal menu claims in claims.md. The owner connects that agent with owner-authorized credentials. The agent proposes the section. The owner approves the exact revision, and the publication updates the same handle. A second agent with read-only access to the public record sees the new signed revision on its next fetch. A third, unauthorized agent that finds a write endpoint in some document cannot publish anything.