How an agent reads a BCP.
Discovery.
Check the brand's domain at /.well-known/brand.md. The root may be a full package or a thin pointer to a canonical Registry URL declared in frontmatter (canonical_bcp, registry_mcp, registry_handle).
Registry HTTPS.
https://registry.encodedbrands.ai/{handle}/.well-known/brand.md and the daughter paths listed in the root. Rendered HTML, JSON, and raw files refer to the same publication. The raw file is the signed source bytes.
Registry MCP.
A handle-bound public endpoint exposes get_bcp. The default returns the root plus available_paths. Pass path for one daughter file, or full_tree: true when every file is needed.
Response shape.
Dated sample from Encoded's own live get_bcp response on 2026-11-08. It is an abridged example, not a schema contract.
{
"trust_level": "verified",
"official_brand_source": true,
"verification_method": "dns_txt",
"verified_domain": "encodedbrands.ai",
"verification_last_checked_at": "2026-11-01T03:17:52.863Z",
"verification_expires_at": "2027-09-01T05:15:48.349Z",
"certification_status": "not_certified",
"integrity": { "manifest_sha256": "…", "manifest_signature": "…", "manifest_signature_kid": "v1" },
"files": [{ "path": "/.well-known/brand.md", "sha256": "…", "signature": "…" }]
}Handling partial reads.
Check response_scope and content_complete. If the agent fetched only voice.md, it has not read the claims. Agents should not state they followed sections they did not retrieve.
Hypothetical example. An agency agent drafting product copy for Northwind Coffee, a fictional brand, fetches claims.md, sees "ethically sourced" marked conditional with a required certifier name, and includes the caveat. It fetches boundaries.md before choosing a placement and finds a hard no on an adjacency. Two fetches, two errors avoided before a human reviewer sees the draft.