Browse documentation
Agent-first documentation · 2026-08-17.4 · Updated 2026-08-17 View as Markdown

Documentation changelog and current guidance

Use the newest documentation version and current generated schemas. Older examples are historical context, not authority when they conflict with the current corpus.

Current version: 2026-08-17.4

Repaired the first live claude.ai OAuth and Ideas List pagination paths:

  • preserved the exact same-origin /oauth/authorize request through Domino phone sign-in instead of falling back to /feed;
  • increased the default non-SMS Ideas List page from 10 to 25;
  • exposed returned, total, and continuation state directly in Ideas List results;
  • made partial-page text say that more lists exist and tell MCP clients to continue with Domino's snapshot handle and opaque cursor;
  • kept the signed snapshot/cursor contract for accounts with more than 25 matching lists.

Preferred guidance:

  • After signing into Domino from Claude, continue to the consent screen and then return to Claude's exact callback.
  • When a user asks for all Ideas Lists, continue until has_more is false and never present a page count as the total.
  • Treat 2026-08-17.4 behavior as current unless a newer docs version is available.

Version 2026-08-17.3

Added standards-compliant OAuth for claude.ai hosted custom connectors:

  • published RFC 9728 protected-resource metadata, OAuth authorization-server metadata, and WWW-Authenticate discovery challenges;
  • added Claude-compatible Dynamic Client Registration with exact current claude.ai and transition claude.com callback allowlisting;
  • added authenticated browser consent with PKCE S256, exact redirects, state, one-time server sessions, CSRF validation, scoped short-lived access tokens, rotating refresh tokens, replay shutdown, and revocation;
  • kept static Agent Access bearer tokens working for Claude Code, Capability API callers, and other header-capable clients;
  • bound OAuth access tokens to the MCP resource and the selected Domino account, client, and abilities;
  • added editable and revocable OAuth Connections to Agent Access plus secret-free audit events;
  • clarified that public docs access, claude.ai OAuth, Claude Code static headers, and Claude API programmatic token use are different paths.

Preferred guidance:

  • Add https://letsdomino.io/mcp in claude.ai, choose Connect, authorize the smallest abilities in Domino, then verify authenticated initialize, tools/list, and a harmless read call.
  • Do not create or paste a static token into claude.ai's hosted connector flow.
  • Keep using static Agent Access tokens for Capability API or Assistant API access.
  • Treat a configured connector as unverified until authentication, tool discovery, and an account-grounded read succeed.
  • Treat 2026-08-17.3 behavior as current unless a newer docs version is available.

Version 2026-08-17.2

Made public documentation discoverable without JavaScript:

  • added server-rendered homepage content with ordinary links to the human and AI-readable documentation;
  • added Markdown content negotiation at the homepage;
  • added a generated sitemap covering public pages, scoped LLM indexes, and every canonical documentation page;
  • advertised the sitemap from robots.txt and the homepage document head;
  • added server-rendered homepage title, description, and canonical metadata;
  • made /docs the canonical human URL for the documentation index.

Preferred guidance:

  • Start automated documentation discovery at https://letsdomino.io/llms.txt.
  • Start human documentation discovery at https://letsdomino.io/docs.
  • Treat 2026-08-17.2 behavior as current unless a newer docs version is available.

Version 2026-08-17.1

Made /mcp the canonical remote MCP endpoint:

  • updated Agent Access settings and public MCP setup examples to advertise https://letsdomino.io/mcp;
  • retained /mcp/v2 as a temporary compatibility alias for previously configured clients;
  • kept both endpoint paths on the same authentication, authorization, and rate-limit contract.

Preferred guidance:

  • Configure new MCP clients with https://letsdomino.io/mcp.
  • Treat /mcp/v2 as a compatibility path, not a separate MCP or capability protocol version.
  • Treat 2026-08-17.1 behavior as current unless a newer docs version is available.

Version 2026-08-12.2

Added idea provenance as a first-class concept:

  • distinguished Domino-suggested ideas from user-originated ideas;
  • explained why “Domino-suggested” is clearer than “Domino-generated” for verified real-world places and events;
  • separated an idea's origin from its structure, user state, and planning state;
  • clarified that research, enrichment, deduplication, or a match to an existing place or event does not erase user origin;
  • routed “Does Domino suggest ideas, or can I add my own?” to the shared concept guide;
  • aligned browsing, Created Ideas, web capture, SMS, MCP, API, and current-boundary guidance with the same terms.

Preferred guidance:

  • Use Domino-suggested for an idea Domino surfaces from its inventory and current context.
  • Use user-originated when a thought or source entered Domino through the user; use Created Ideas when naming that product section.
  • Explain provenance only when it helps the user. Do not add taxonomy to an otherwise simple task answer.
  • Treat 2026-08-12.2 behavior as current unless a newer docs version is available.

Version 2026-08-12.1

Made idea capture a complete first-class documentation path across every supported surface:

  • added a canonical explanation of source preservation, enrichment, result types, destinations, singular capture, and roundups;
  • replaced generic web directions with the exact Create → Create idea → Save idea flow and current review controls;
  • aligned Created Ideas guidance with the current capture-first creation experience;
  • added SMS examples for text, links, MMS, people/list context, follow-up corrections, later artifacts, planning intent, and mixed roundup outcomes;
  • added complete MCP and Capability API recipes for singular create, polling, clarification, terminal proof, roundup scope, destination, selection, retry, archive, and stale generations;
  • documented the opaque media-reference boundary and safe web/SMS handoff;
  • added capture intents and search keywords to machine-readable discovery.

Preferred guidance:

  • Start with How Domino idea capture works when the user is asking what the feature can do.
  • Fetch the dedicated web, SMS, MCP, or API capture guide before giving surface instructions.
  • Treat every accepted capture and roundup mutation as non-terminal until the paired read confirms saved state.
  • Treat 2026-08-12.1 behavior as current unless a newer docs version is available.

Version 2026-08-11.3

Expanded the agent-first corpus from basic task coverage to the full customer planning lifecycle:

  • direct and idea-based plan creation;
  • RSVP deadlines, tipping, capacity, delivery, active edits, roster management, plan progress, discard, and cancellation;
  • recipient authentication, invite claiming, RSVP changes, and calendar handoff;
  • Connected Friends, Friends in Mind, connection links, and availability boundaries;
  • complete My Dominos status and personal-inventory guidance;
  • detailed singular capture and roundup recovery;
  • recipient-side shared Ideas List copying and connection consequences;
  • created ideas, private places, account access, and security;
  • Agent Access token creation, least privilege, rotation, and revocation;
  • explicit ChatGPT, Claude, and static-bearer MCP compatibility checks;
  • SMS, MCP, and Capability API task recipes;
  • cross-surface support and current-limits references;
  • machine-readable documentation manifest, search, word counts, and token estimates;
  • semantic coverage assertions for representative support questions.

Preferred guidance:

  • Use task-specific pages rather than only /llms.txt.
  • Use /docs/manifest.json for catalog metadata and /docs/search.json?q=... for discovery.
  • Use authenticated MCP tools/list, /docs/mcp-tools.json, and /docs/openapi.json for current schemas.
  • Treat 2026-08-11.3 behavior as current unless a newer docs version is available.

Version 2026-08-11.2

Added complete provider-specific calendar guidance:

  • direct Google authorization and access levels;
  • Apple/iCloud public calendar publishing;
  • Microsoft Outlook/Microsoft 365 ICS publishing;
  • other ICS and webcal providers;
  • calendar privacy, secret-link handling, multiple calendars, management, and troubleshooting.

Preferred calendar guidance changed from “Google or a calendar URL” to explicit provider routing with accurate Apple and Microsoft limitations.

Version 2026-08-11

Created the initial agent-first documentation system:

  • stable Markdown task pages;
  • HTML/Markdown content negotiation;
  • global and surface-scoped llms.txt indexes;
  • llms-full.txt offline corpus;
  • generated OpenAPI and MCP tool catalogs;
  • generated capability reference;
  • cross-surface outcome and confirmation rules.

Deprecation policy

When guidance changes:

  1. update the task page and documentation version;
  2. update generated schema descriptions when the runtime contract changed;
  3. record the preferred replacement here;
  4. keep old behavior only when a real supported compatibility window exists;
  5. add a semantic documentation test for the user question that exposed the change.

Assistants should not follow a deprecated menu path, capability version, authentication mechanism, or confirmation rule merely because it appears in an older answer.

Machine checks

Use:

  • X-Domino-Docs-Version to identify the corpus version;
  • ETag to cache exact Markdown safely;
  • X-Domino-Docs-Words and X-Domino-Docs-Token-Estimate to budget retrieval;
  • documentation_fingerprint in generated machine references to detect contract changes.

LLM answering guidance

  • Prefer the newest fetched page over model memory.
  • When a user reports different visible controls, acknowledge possible rollout drift and reason from their screen without inventing hidden behavior.
  • Never combine an old confirmation rule with a new runtime contract.