Everything, in one place.

Download the installer for macOS, Windows, or Linux from your dashboard after signing up. Run it once — it drops the app in your applications folder and adds a small icon to your system tray. No admin rights needed, no OS surgery. Uninstall from your OS's normal apps list any time.

Start with one specific project folder — not your home directory. The dashboard's inline path validation nudges you toward safe choices (green tick for a real folder, red hint for a typo). You can add, remove, or re-label folders any time; the next request from your AI sees the change.

From your dashboard, browse the MCP catalog, click 'Add' on the one you want, and scope it (which folders it can reach, for filesystem-style MCPs). Your desktop app picks up the change automatically. Removing is equally quick — one click, and the MCP is gone from your connector.

No. The desktop app opens an outbound tunnel to us — same as any browser or messaging app opening a connection. Nothing incoming, no firewall rules, no port forwarding.

Yes. Each machine gets its own instance and its own connector URL. From the same account you can have, say, one client on your home workstation and one on a small always-on box behind a shop counter — each independently configured.

The Model Context Protocol — a common language AI apps use to safely reach files, tools, and data outside themselves. Anthropic created it for Claude, but many AI apps now speak it (ChatGPT, Gemini, LibreChat, agentic coding tools, and more). If your AI supports MCP, this platform is how it plugs into your machine.

Only the response to a specific request your AI made — a file's contents when it asked for it, a directory listing, a command's output. Never a whole folder, never on a schedule, never as a background sync. We don't store, cache, or mirror your files anywhere.

Your desktop app opens an outbound TLS tunnel to our server the moment it starts, using your unique tunnel token. Requests from your AI arrive at our server, get authenticated per-user, and route into your tunnel only — never anyone else's. When you close the desktop app, the tunnel closes; no request can reach your machine until you open it again.

No. Every account gets its own encrypted tunnel and its own set of tokens. Requests are authenticated at multiple independent layers — dashboard login, desktop device access, AI connector access — and the routing is per-user by design. A leak in one flow cannot expose your files to another user's requests.

The tunnel closes immediately. Any incoming request from your AI gets a clean error — nothing reaches your machine until you open the app again. It's the same reason we run in your system tray: opening and closing the app is your 'on/off' switch for reachability.

In Haarlem, North Holland, Netherlands. EU data residency by default; no cross-border transfer concerns for this infrastructure.

The client is the desktop app on your machine. It can host one or more MCPs (Filesystem, Shell, GitHub, …), each configured with its own settings and — for filesystem-style ones — its own scoped folders. All the MCPs on one client share a single connector URL that you paste into your AI. So: one machine → one client → many MCPs → many folders.

A connector URL is what you paste into your AI's integration settings (Claude Settings → Integrations → Add custom connector, or the equivalent for other AIs). It's the address your AI uses to talk to the MCPs on your machine. The URL is per-client-per-machine — one URL brings in every MCP configured on that client, each showing up as its own tool group in your AI.

Because a client can host multiple MCPs at once (Filesystem + Shell + GitHub, say), and each MCP already gets its own internal route inside the shared tunnel. That means multiple tools share one secure connection — cleaner networking, less overhead, same per-user isolation. The security guarantee is per-user, not per-MCP.

Yes. Each machine has its own client instance and its own connector URL. You can paste both URLs into your AI so it sees both machines' tools, or keep them separate. This is how multi-machine setups work today — the Connectors dashboard lists the per-machine URLs.

Up to 2 MCPs configured, 100 tool calls per day, and every security guarantee (same encrypted tunnel, same per-user isolation). Free forever, no card required. If that's enough for how you work, stay on it.

When you need more than 2 MCPs, or your usage regularly hits the 100/day cap. Pro is €9/month with a much higher daily quota and no cap on how many MCPs you configure.

Pro runs on our shared gateway — very fast, per-user tunnels keep you isolated, but the compute is shared. Isolated gives you your own dedicated instance we host and manage, region-pinned (EU or US), with a much higher daily quota and audit-log export. Same security guarantees on both.

The Platform tier is roadmap work — team accounts, bring-your-own-auth, self-hosted deploys. Ping us via the contact form if that's what you're looking for so we can prioritise. Until then, teams typically start with a shared Pro account or one Isolated instance.

Yes. Upgrade or downgrade from Profile → Plan. Downgrades take effect at the end of your current billing cycle; upgrades are immediate. Data stays put; you never lose configurations by moving between tiers.

Three things, in order: (1) is the desktop app running on your machine? Check the tray icon. (2) Does your machine have internet? (3) Try closing and reopening the app; it reconnects with exponential backoff. If it stays disconnected after all three, the dashboard shows the last reason it saw — check there.

On macOS, check Security & Privacy — allow the app if Gatekeeper blocked it. On Windows, run once from the installer location (not from a downloaded ZIP). On Linux, make sure the binary is executable ('chmod +x'). If it still won't start, our contact form gets your log lines to us fast.

The MCP the folder was added to might not be the one your AI is asking. Check that the folder is on a filesystem-style MCP (not, say, a Shell MCP). Also confirm the folder path is exactly as it appears on your machine — no trailing slash, correct capitalisation on case-sensitive filesystems.

Requests beyond the cap get a 429 response with a Retry-After header pointing at midnight UTC. Your desktop app and dashboard keep working normally — you just can't fulfill new tool calls until the cap resets. Upgrade to Pro or Isolated any time to raise the ceiling.

Usually a flaky network on the machine side. The app auto-reconnects with exponential backoff — a burst of quick reconnects usually settles into a stable one within a minute. If it persists, check whether a corporate VPN or firewall is dropping long-lived connections.

Profile → Plan. Upgrades are immediate; downgrades take effect at the end of your current billing cycle so you don't lose access mid-month. Data stays put across changes.

Paying accounts get invoices via Profile → Billing, retrieved live from Stripe. Free tier has no invoices to show.

Profile → Billing → Update payment method. Takes you through Stripe's secure form. We never touch your card details ourselves.

Profile → Account → Delete account. Confirms once, then deletes your dashboard data, revokes all tokens, and closes the account. No hidden retention.

The desktop app is configured to launch on login (per your OS's normal autostart). Once it starts, it reconnects the tunnel automatically. If autostart is off, you'll need to open the app manually before requests can reach your machine.

The tunnel drops. Any in-flight request from your AI gets a clean error; your AI sees a 'connector unreachable' message. When your connection comes back, the app reconnects automatically. Nothing queues; requests are point-in-time.

Same experience as above — the tunnel drops, your AI can't reach your machine. Our uptime dashboard is on the roadmap; for now, the contact form is the fastest way to know if what you're seeing is us or you.

The tunnel closes cleanly. If autostart is configured (default on macOS and Windows), the OS relaunches the app; on Linux, a systemd user unit does the same when set up. Recovery is silent from your AI's perspective — the next request just works after ~30 seconds.