The Cloud (thecloud.so) is a workspace any AI can read and write — over the Model Context Protocol (MCP). Point your assistant at the endpoint below and it uses The Cloud as durable, shared memory: pages, tasks, calendar, and more. No insider setup required.
One HTTP endpoint, JSON-RPC 2.0. This is the only address you need:
https://mcp.thecloud.so/mcp
Every served Studio app inherits the same credentialed session probe. Anonymous is an honest 200 response, so storefronts and headless checks never need an interactive OAuth attempt just to learn the current state.
fetch("https://lovable-crab-328.convex.site/session/whoami", { credentials: "include" })
// signed in: {"signedIn":true,"name":"…","member":{…}}
// anonymous: {"signedIn":false}Iframed apps can send the platform shell {"type":"cloud:identity:request","app":"<appId>","v":1} and receive {"type":"cloud:identity","signedIn":true,"name":"…"}. Replies go only to direct child frames and use the requesting frame's exact origin as targetOrigin.
One step per client. Your browser opens once to sign in to your Cloud account, and the connection is live. No secret is copied, pasted, transcribed, or stored anywhere.
Claude Code (CLI)
One terminal command. Claude Code drives the browser sign-in.
claude mcp add --transport http the-cloud https://mcp.thecloud.so/mcp
Cursor
Add to CursorOne-click install with the button below, or paste the config into mcp.json + Reload Window. First connection opens the browser to sign in.
File: ~/.cursor/mcp.json (global) or project .cursor/mcp.json
{
"mcpServers": {
"the-cloud": {
"url": "https://mcp.thecloud.so/mcp"
}
}
}Gemini CLI
One terminal command. Gemini CLI opens your browser to sign in. There is no JSON file to edit and no API key.
gemini mcp add --transport http the-cloud https://mcp.thecloud.so/mcp
Claude Desktop
Settings → Connectors → Add custom connector. Paste the endpoint. Sign in when prompted.
https://mcp.thecloud.so/mcp
For environments where the browser sign-in isn't available (continuous integration, scripted setups), or clients that don't support MCP's OAuth flow yet (Windsurf, ChatGPT), generate an API key under Settings → AI and pass it as a Bearer header:
{
"mcpServers": {
"the-cloud": {
"type": "http",
"url": "https://mcp.thecloud.so/mcp",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}
}The in-app setup guide under Settings → AI has the exact snippet for each client, including Windsurf and ChatGPT.
Once connected, your assistant can read and write your Pages and Memory, manage Tasks and Calendar events, and reach your Files and Profile — using The Cloud as the memory it carries between sessions.
The Cloud exposes the workspace as 618 MCP tools across 32 families, grouped by what they reach. Read tools return your data; write tools create or change it and hand back a link. This list is generated from the live tool registry — the same source as llms-full.txt — so it never drifts from what the server actually serves.
Durable cross-AI memory: read, search, write, and organize what every AI should carry between sessions. What an agent writes lands proposed and waits in the owner's triage queue before it counts as verified.
The identity ledger: resolve-or-create nonfungible Things (people, repos, products, models, places) and record append-only Facts about them with provenance.
Rich documents and the Spaces holding them: create, read, edit, restructure, tag, and publish.
To-dos, task lists, and the daily page: the user's plan for today and what is outstanding.
Events, availability, free-time search, and public booking types.
The Timer surface as verbs: start a Timer, Stopwatch, or Pomodoro, pause and resume it, re-time it without restarting, rename it, stop it (logged) or cancel it (not logged), and read both what is running and the Log of what was.
Workspace files and folders, plus the searchable index of attached drives.
MCP-native spreadsheet/database grids: columns, rows, queries, summaries, and sharing.
Form definitions, fields, capacity, notifications, SEO, and collected responses.
People, contact folders, lead scouting, and drafted outreach to a contact.
Connected Gmail accounts: search, read, send, label, archive, and drafts.
Audience campaigns, deliverable packages, send readiness, newsletters, and CAN-SPAM compliance settings.
Connected Telegram chats: search, sync control, quote mining, and contact mapping.
Ora in the room: read a Space's bridged Slack channels, threads, and history, and post as Ora through the per-class autonomy dial and approval queue.
Connection status, posting, and archiving.
The content calendar a Space plans on: draft, review, approve, schedule, and publish posts. The agency drafts and schedules; the client approves; nothing reaches a channel unapproved.
The workspace music library and connected Spotify playlists.
Real multi-file apps built inside the workspace: files, manifests, secrets, rendering, sharing, and verification.
Image and video generation, transcription, conversion, compression, clip styling, blurring, and stream-safety review.
Handwriting-to-font creation from a template photo, uploaded font families, applying a font to an appearance slot, and exporting the packaged files.
Dispatch agents to your own attached hardware, watch runs, and collect results.
Drive a headless browser on your own attached machine: open a page, read it as markdown, screenshot it, act on it (consent-gated), and close it. Public web only in P0.
Scheduled recurring agent runs: create, customize, start, pause, and run now.
Repos, issues, pull requests, code search, and the read/branch/commit path to opening a PR.
Storefront catalog, products, variants, inventory, and Amazon product resolution.
Public profile pages, link and text blocks, and visibility.
The workspace app launcher plus brand-mark and character-asset lookup.
Bring an existing Notion workspace into The Cloud.
The consent floor as tools: record and revoke consent, mint and revoke scoped access, and read the telemetry ledger.
Append-only provenance + credits accounting: register signed contributions against artifacts, walk their lineage, and read the credits they earned.
Account provisioning, invites, signup approval, and staged account content.
The connected Whop account through the generalized credential vault: connection status, allowlisted reads (memberships, companies, plans), and confirm-gated writes under the hardened read-allowlist doctrine.
Every tool carries a safety annotation — reads, ordinary writes, and destructive actions are each marked, so your client knows what is safe to run and what to confirm first.
Everything a connected AI creates is Private by default. Widening exposure — sharing, publishing, or any change that makes something more visible — and every destructive or irreversible action (delete, send) needs your explicit, per-action approval. An AI cannot self-approve: it can request the change, but it only takes effect once you confirm it in The Cloud. The safety annotations each tool carries — read, write, destructive, consent, and external — are how your client knows which is which before it calls anything.
The connector won't connect
Confirm you pasted the exact endpoint above. Remove the connector and add it again: the browser sign-in runs once more and the session refreshes.
An “unverified app” screen when connecting Gmail or Calendar
Google verification for those scopes is still in progress. Until it finishes you can proceed past the screen, or skip the Email and Calendar tools. Every other tool works without them.
An API-key client returns 401 Unauthorized
The key is missing or was revoked. Generate a fresh one under Settings → AI and update the Authorization header in your client's MCP config.
Tools look missing or stale after a change
Reconnect the server, or restart your client, so it re-reads the tool list from a fresh session.
Once connected, you talk to your AI the way you always do — it reaches The Cloud on its own. A few things to try:
Remember that the directory submission is the priority for this release.
The AI writes the fact into your VastMemory. Every later session that connects The Cloud carries it, so you never re-explain.
Draft a one-page brief for the Q3 roadmap and put it in my Work space.
The AI creates the page in The Cloud, writes the brief, and hands back a link you can open or share.
What do I have on tomorrow, and what's still open on the launch?
The AI reads your calendar and tasks and answers from your real data, not a guess.
Inside The Cloud, a connected AI takes on the role of Ora — The Cloud's one governed AI. Ora is a role, not a disguise: your assistant stays your own model on your own surface, and stays honest about it. Every artifact it authors is signed with all three — role, model, and surface (for example Ora · Claude Opus · Claude Code) — so provenance stays exact. Whether you connect your own AI or use the Ora built into The Cloud, both work from the same memory under the same rules.
On connecting, a connected AI pins The Cloud's Charter — the master document that keeps it guided, governed, and in sync with your memory for the whole session. Sharing and visibility always stay under your control: an AI never makes anything public on its own.