Your Oesnada.
Inside your AI.

01

Connect once. Keep your shelf.

Relay speaks the open Model Context Protocol at /api/mcp. Claude Desktop, ChatGPT remote MCP, and other compatible clients can connect to the same Oesnada layer without a bespoke plugin or a new database of your reading life.

02

Ask the thing you actually care about.

Relay turns natural questions into scoped MCP calls. Ask for a catch-up, a topic line, a source trail, or why a story was selected. Oesnada returns structured context; your model writes the final answer in the client where you asked.

03

The toolbelt is the product surface.

Relay exposes small tools with clear jobs, not one giant text dump. Read tools stay read-only, return bounded context, and are built for questions a reader naturally asks.

• catch_me_upRecent selected stories plus reader profile, Library, and Signal Memory context.
• ask_my_oesnadaAnswers a reader question from saved stories, memory, and the current personal feed.
• search_my_librarySearches saved, liked, and read stories in the reader Library.
• get_topic_timelineBuilds a topic, source, company, person, or recurring-line timeline from selected stories.
• explain_why_selectedExplains why Oesnada selected a story and how it connects to the reader line.
• get_story_contextReturns one story with structured Oesnada context for deeper work.
• get_personal_feed / get_saved_stories / get_signal_memoryPrimitive shelves for clients that want feed, saved items, or aggregate memory directly.

04

It can write back, when you say so.

Relay can also turn a conversation into Library and Signal Memory movement. These tools require write scopes and only change state when the client asks for that action.

• save_storySaves a story to the Library and records it as a strong interest signal.
• deepen_storySaves, marks read, and records a medium-strength Signal Memory event.
• follow_signalTells Oesnada to bring more of a topic, source, category, lens, or story pattern.
• mute_signalTells Oesnada to bring less of that same kind of signal.
• record_interest_signal / reduce_interest_signalCompatibility tools for explicit memory edits from older clients.

05

Resources and prompts come with it.

Clients that prefer MCP resources can read oesnada://reader/profile, oesnada://reader/library, oesnada://reader/signal-memory, and oesnada://story/{id}. Prompt-capable clients get oesnada_catch_me_up and oesnada_deep_signal_brief as ready-made starting points.

06

Setup lives inside Library.

Open your Library, connect Signal Memory to AI, then choose the path your client understands: one-command Claude Desktop setup, remote MCP connector URL, or the advanced manual token view for debugging. Tokens can be revoked from the Library when the line should close.

• 1. Open LibraryThe MCP panel lives with your saved stories and Signal Memory controls.
• 2. Connect your clientUse Claude Desktop setup or point any remote MCP client at /api/mcp.
• 3. Keep controlBearer tokens are scoped, visible only when needed, and revocable from the same panel.

07

The permission model is boring on purpose.

Relay uses scoped bearer tokens and OAuth discovery metadata. Read paths require read:profile, read:library, or read:signals. State-changing paths require write:libraryor write:memory. Nothing needs broad open-world access to your account. For the broader data policy, see privacy.