2026 10 10 connector submission packet - SyberLabs/RISE GitHub Wiki

RISE connector submission packet

Filed 2026-10-11 from the paid account, live build a7387e6d; Community listing pending the automated scan. The portal had no field for the screenshots or the allowed link URIs; they are kept ready in docs/assets/directory/ and under Server below. Recorded in the readiness register §7 and in LIVE-004.

Date: 2026-10-10. For: Mateo (SyberLabs), who files it. Source: the readiness register (§5 decisions, §7 checklist, §8 listing copy, §9 questions) and the directory screenshots.

Every field Anthropic's portal asks for, filled, in the order to paste it. The listing copy is §8 of the register as drafted; the decisions behind it are the owner's of 2026-10-10: submit without the Plus voice, terms corrected and the test payment link removed, the demo prompts verified through Claude, the MCP Inspector run recorded, and the owner submits from the paid account that will own the listing.

Before you start

The freeze (2026-10-11). Connector-facing code is frozen from the build this packet names until the submission is filed. The build to submit is the one release.txt shows at filing time; record it in the register (§7) and in LIVE-004. Nothing merges to the card, the Worker's MCP route, the tool text or the pages until then. The last connector change was #631 (a613d0e7): the bed fades out when the voice has ended.

What must be true before you press Submit. All but the last are yours, in this order.

A fresh chat, first thing. Start a new chat before any other RISE use that day, with the connector added (https://rise.syberlabs.io/api/mcp), and ask one demo prompt. The card must appear without a page reload, through the permission prompt. This is the one check to hold the submission for if it fails: it is a reviewer's first interaction, and the "first chat of the day" blank card (LIVE-016) has never been reproduced or fixed.
One long reading, played through. A lesson of twenty or more beats with figures and holds (the potato lesson is the right shape), played start to end without seeking, on your Edge voice. The sixteen-beat defect (#629) lived three days because every fixture was short. Then copy "About this reading": it should show no run.failed and no refused: line.
The demo prompts, each once, in the claude.ai chat and in the Claude iOS and Android apps. Each should show the card; press Play once in each. The tool calls are already proven (the Inspector run, the Claude run); what only you can see is the card rendering and playing in each app. The attestation below claims all three.
The negative prompts. "Book me a table for two tomorrow night." and "Generate a picture of a cat." must not call RISE. The listing's health badge counts refusals as errors, so a tool called wrongly costs the listing over time.
Sign in with the paid Claude account that will own the listing. The listing belongs to the organization it is submitted from.
The live site matches this packet: /terms names no price, /api/plus/status says available: false and /api/plus/config serves no payment link, /.well-known/security.txt names the policy. Re-checked on 2026-10-11 against a613d0e7; the terms' planned-price sentence found then is removed in the same pull request as this revision.

Server

Field	Value
Connector type	MCP connector (remote, Streamable HTTP)
Server URL	`https://rise.syberlabs.io/api/mcp`
Authentication	None
Before connecting, users need	Nothing: no account, no key, no payment
Reads or writes user data	Neither. Both tools are read-only (`readOnlyHint: true`)
API	Our own: a Cloudflare Worker at `rise.syberlabs.io`
Allowed link URIs	`https://syberlabs.io` and `https://rise.syberlabs.io`

Listing

Name: RISE

One-liner: Turns an answer into a short reading you watch and listen to: your device speaks it, the words appear as they are spoken, and a drawn visual plays behind them.

Description:

Ask Claude to explain something and present it in RISE. Claude writes the answer as a reading: a title and its passages (up to 64 beats and 20,000 characters, which is about twenty minutes at the longest; most readings are a minute or two), and, if it likes, a colour theme, a look, and a diagram or figure of its own. RISE shows it in the chat as a card. You press Play. Your device's built-in voice reads it aloud, the words appear as they are spoken, and a visual plays behind them.

You stay in control: play, pause, go back or forward a passage, hear a passage again, change the pace, and in Settings choose the theme, how vivid the imagery is, whether it moves at all, the text size, the sound and the voice. If your device has no voice, the reading still plays, paced, with every word shown.

The imagery is drawn, not generated. RISE draws it live with its own procedural engines: line drawings, a strange attractor, soft light, fractal flames, spectral plates. When Claude adds a diagram or figure, RISE checks it first and runs it apart from the page, with no network and no access to your data. RISE calls no image, video or speech model of its own: the voice is your device's built-in speech.

RISE needs no account and no key, and it costs nothing. It receives only the reading Claude writes for it, plays it, and keeps none of it. It never sees your conversation, memory or files.

What it does not do: it takes no actions, makes no pictures to order, and is not for tables, code, or long reports. Claude answers those itself.

Made by SyberLabs. Privacy: rise.syberlabs.io/privacy

Field	Value
Category	Education (primary); Media or Learning if offered
Icon	`public/android-chrome-512x512.png` in the repository (512 × 512 PNG), also at `https://rise.syberlabs.io/android-chrome-512x512.png`
Documentation URL	`https://rise.syberlabs.io/apps`
Privacy policy URL	`https://rise.syberlabs.io/privacy`
Terms URL (if asked)	`https://rise.syberlabs.io/terms`
Support contact	`[email protected]`
Security contact (if asked)	`https://github.com/SyberLabs/RISE/security/advisories/new`, policy at `https://github.com/SyberLabs/RISE/blob/main/SECURITY.md`

Test and launch instructions

Paste this into the reviewer-instructions field:

RISE needs no account, credentials or payment; there is nothing to sign in to. Add a custom connector with the URL https://rise.syberlabs.io/api/mcp (no authentication). Then ask, for example, "Show me the parts of an animal cell in RISE." Claude calls rise_guide for the format reference when it writes in a named style, then rise_present with the reading. The card appears in the chat showing the title and Play; press Play. Your device's built-in speech reads it and the words appear as they are spoken; if the device has no voice, the reading plays paced with every word shown. Settings, on the card's bar, changes the theme, imagery, motion, text size, sound and voice. rise_present refuses a malformed reading with a one-line message, and Claude can correct it and call again. Both tools are read-only and store nothing.

Compliance answers

Question	Answer
Financial transactions	None. Nothing in RISE is for sale.
AI media generation	RISE calls no media model. The voice is the device's own speech synthesis (Web Speech API); diagrams and figures are code or SVG the model writes, rendered by RISE.
Conversation data	Only the tool's argument, which is not stored. Cloudflare's request logs keep metadata (method, path, status, timing, the client address Cloudflare sees) for at most 7 days, as the privacy page states; no tool text is logged.
Prompt injection	The tool returns a one-line receipt or a clipped validator message; `rise_guide` returns a format reference, not instructions.

The AI-media answer is the register's §8 wording without its internal note. That note stands: if Anthropic reads on-device speech as AI audio, that is question 1 of the register's §9, to ask at [email protected]. The Plus voice is not part of this submission (register §5.1, decided 2026-10-10): it is unreachable from the Claude card and named in no connector text.

Screenshots

Upload in this order from docs/assets/directory/, each with its prompt in the prompt field. All five are PNG, 1140 × 753, the card alone.

File	Prompt to paste
`01-sky-scene.png`	Give me a RISE reading on why the sky is blue, in the premium-educational style.
`02-animal-cell.png`	Show me the parts of an animal cell in RISE.
`03-garden-reflection.png`	Present a short reflection on Marcus Aurelius and the inner citadel in RISE, in the Garden look.
`04-settings.png`	Explain how black holes bend light, and present it in RISE.
`05-poster-play.png`	Show me what a derivative is, as a RISE reading in the premium-educational style.

Demo prompts

"Explain how black holes bend light, and present it in RISE."
"Show me what a derivative is, as a RISE reading in the premium-educational style."
"Give me a RISE reading on why the sky is blue, in the premium-educational style."
"Present a short reflection on Marcus Aurelius and the inner citadel in RISE, in the Garden look."
"Show me the parts of an animal cell in RISE."

Prompt 5's tool calls are recorded in the Claude run. The others are verified by your own runs under "Before you start"; note each outcome in that record when done. If a prompt fails, drop it here before filing (the portal needs at least three).

If the portal asks for cases where the connector should not be used: "Book me a table for two tomorrow night." "Generate a picture of a cat." "Write a 10,000-word report and read it in RISE."

Attestation

The text you tick:

Every tool was run in MCP Inspector and as a custom connector in Claude, including the Claude iOS and Android apps.

The evidence:

MCP Inspector 2.10.1 against the live server: tools/list, both tools called, refusals, resources/list, resources/read of the card. INSPECTOR-RUN-2026-10-10.md
The custom connector called from Claude: rise_guide and rise_present with the worked parts-of-a-cell, accepted. CLAUDE-RUN-2026-10-10.md
The claude.ai chat, iOS and Android runs: yours, from "Before you start" above. To be added to the Claude run record when done. Do not tick the attestation before them.

File it

Go to claude.ai/directory/manage, signed in with the paid account that will own the listing.
Choose MCP connector.
Paste, field by field:
- Server URL, authentication (none), and allowed link URIs from Server.
- Name, one-liner, description, category, icon, documentation, privacy and support from Listing.
- The reviewer instructions from Test and launch instructions.
- The four answers from Compliance answers.
- The five images with their prompts from Screenshots, in order.
- At least three of the Demo prompts that passed.
Tick the attestation only once the three runs in Attestation are all done.
Submit. The automated scan runs first; the listing is Community by default. Review questions go to [email protected], MCP App questions to [email protected].
Record the submission date and anything the portal asked that this packet did not cover in the register (§7) and in LIVE-004.

Published from docs/product/discussions/2026-10-10-connector-submission-packet.md. Edits made here are overwritten on the next push to main.