Full Claude Code sessions running directly on your managed servers, in the directory where the code actually lives, streamed to your browser. No repo syncing, no file copying, and no inbound ports opened on anything.
Remote Claude came out of our own workflow, not a roadmap. Remote desktop is the right tool for plenty of jobs, and we still use it for most of them. What we wanted was something purpose-built for one specific case: a whole team running AI coding sessions against a single shared development server, all day, from whatever device happens to be in front of them.
So everyone signs in through the RMM in a browser and works against the same development server. Each developer owns a different module — billing, reporting, integrations, and so on — so we're rarely editing the same files at the same time.
Working with Claude turns out to involve a lot of images: error dialogs, screenshots of a broken layout, a diagram of what the flow should be. Making that a single paste into the terminal, with the file landing on the server ready to analyze, mattered more to us day to day than almost anything else on this page.
Several developers working the same codebase concurrently, each in their own session, without queuing for a seat on the server.
Because everyone is on one development server rather than their own copy, each person's changes show up for the rest of the team right away.
No syncing step between where the code is written and where it runs, which takes a whole class of delay out of the loop.
Shipping sooner means customer and field-user feedback comes back sooner, and it lands while the change is still fresh.
A browser tab per developer, with paste and image attachments handled natively and no per-seat session planning to think about. Narrow by design, and a good fit alongside the remote access tooling you already run.
Built for our own developers first. Everything here is what we run day to day.
Claude Code runs on the managed server itself, against the real working directory. Nothing is staged, mirrored, or shipped elsewhere first.
Sessions start in the directory you point them at, on the machine that already holds the project. No repo syncing and no file copying to get started.
Because the session executes locally on the server, your source stays on your infrastructure for the whole session.
A live terminal is streamed straight to the browser. Any machine with a browser is a working terminal, with no agent or client tooling to deploy first.
Live streamed sessions on the server itself. No SSH client, no VPN profile, no local setup.
The agent reaches out. Nothing reaches in. There is no listening service to find, which means there is nothing for a port scanner to report.
No inbound SSH, RDP, or VPN is required to reach a session. Nothing has to be opened on the managed server for one to run, so the machine stays invisible to port scanners.
The agent establishes the connection from the inside out. That keeps the firewall posture unchanged and avoids punching holes to reach a server.
All traffic moves over encrypted WebSocket and HTTPS, with sessions authenticated by JWT and scoped to a specific device and tenant.
Multiple projects per device, several live sessions side by side, and enough lightweight project management to keep the work organized without leaving.
Define as many Claude projects per device as you need. Each one is just a name and a working directory, with an optional custom Claude config path.
Run several sessions at once with tabbed navigation, or split into side-by-side panes to work two projects at the same time.
A built-in to-do / in progress / done board plus per-project notes. Lightweight project management that lives inside the workspace itself.
Sessions survive page refreshes, network drops, and closed tabs, reconnecting automatically with scrollback and context intact. Only an explicit End Session stops one.
Paste a screenshot or diagram straight into the terminal, or use the attach button for files up to 7 MB. The image lands on the remote device with the path ready for Claude to analyze.
The microphone works inside the session, so you can dictate to Claude instead of typing. Useful for long prompts, and for the times it is easier to describe the problem than to write it out.
Scoped access, tracked spend, and a complete audit trail. The feature ships behind a per-tenant flag and is off until you turn it on.
Assign specific projects to individual people through the built-in Claude User role, at either use (open and run sessions) or manage (also edit project settings). Scoped users see only what they are assigned, with no device list and no admin screens.
Generate a secure link that lets someone open a Claude session on one project with no account, no login, and no setup. Set an expiry, cap the session count, and revoke instantly, including for sessions already running. Tokens are never stored server-side, only a cryptographic hash.
Every session is tracked automatically, including share-link sessions attributed back to their link. The dashboard covers sessions, input and output tokens, cache utilization, and precise per-model cost across Opus, Sonnet, and Haiku.
Cost and usage split by device, project, and day, with trend charts over 30, 90, or 365 days and CSV export. No shadow AI and no surprise bills at the end of the month.
Session starts and ends, access grants, and share links created or revoked, each recorded with the user, device, and IP behind it.
Sessions are JWT-authenticated and scoped to a device and tenant, and the feature sits behind a per-tenant flag that is off by default. An unauthorized user cannot even discover that a project exists.
Four short screen recordings covering what Remote Claude does, how a team shares a project, and how it is kept contained. Click a thumbnail to play inline.
02:35
Full Claude Code sessions streamed to a browser, running in the directory where the code actually lives. No repo syncing, no file copying, and no dependence on the machine you happen to be at.
02:07
How a team gets into the same project on the same server. Everyone signs in through the RMM, works against shared code, and sees the session from whatever device is in front of them.
02:17
AI coding treated as a managed service rather than a loose tool. Sessions run on infrastructure we monitor, under the same access control and oversight as the rest of the estate.
04:44
The security side of Remote Claude. How access is controlled and kept contained when an AI terminal is running against real infrastructure.
Get in touch to walk through Remote Claude on a live environment, or to talk about how it would sit alongside what you already run.