Anthropic's new feature turns your phone into a window to your local coding agent. No VPN, no port forwarding, no cloud VMs. Your code never leaves your machine.
By LDS Team
February 26, 2026
On February 25, 2026, Anthropic quietly shipped a feature that changes the relationship between developers and their machines. Claude Code Remote Control lets you start a coding session on your laptop, pull out your phone, and keep working from wherever you are -- the couch, a coffee shop, or walking the dog. The local session keeps running. Your phone is just a window into it.
"Take a walk, spot the sun, walk your dog without losing your flow," wrote Noah Zweben, Claude Code's Product Manager, in his announcement on X.
That pitch sounds simple. The engineering underneath is not. Remote Control maintains a live bridge between your local terminal and any device with a browser, while ensuring your source code, environment variables, and MCP servers never leave your machine. It shipped alongside a second feature -- remote MCP server support -- that lets tools like Sentry, Linear, and Figma plug directly into Claude Code without any local server installation.
Both features arrived during a week when Claude Code's numbers tell their own story: a $2.5 billion annualized run rate (more than double since January 2026) and an estimated 4% of all public GitHub commits now authored by the tool.
Here is everything you need to know about Remote Control -- what it is, how to use it, how it stays secure, and what it means for the way developers work.
The core idea: your phone becomes a remote terminal
Remote Control is not a mobile app that runs code in the cloud. That product already exists -- Anthropic calls it "Claude Code on the Web," and it spins up cloud VMs to run your sessions. Remote Control is architecturally different.
When you activate Remote Control, your local Claude Code process registers with the Anthropic API over an outbound HTTPS connection. It never opens inbound ports on your machine. The Anthropic server then acts as a message relay. When you connect from your phone or another browser, the server routes messages between that device and your local session over a streaming connection.
The critical distinction: only two things traverse this bridge -- the messages you type and the tool results Claude generates. Your source code, file system, environment variables, MCP server configurations, and .claude/ project settings all stay on your local machine. Your phone is rendering a remote view of a process that is running on your laptop's hardware, in your laptop's filesystem, with your laptop's credentials.
| Aspect | Remote Control | Claude Code on the Web |
|---|---|---|
| Where code runs | Your local machine | Anthropic cloud VM |
| File access | Full local filesystem | Cloned repository only |
| MCP servers | Your local MCP servers | Not available |
| Setup required | Claude Code installed locally | GitHub account + Claude GitHub app |
| Use case | Continue local work from another device | Start tasks without local setup |
Setting it up takes 60 seconds
Remote Control requires Claude Code version 2.1.52 or later. Check your version and update if needed:
claude --version
claude update
You also need to be logged in with a Claude Max subscription ($100 or $200/month). API keys are not supported -- you must authenticate through claude.ai:
claude
/login
Once authenticated, start a Remote Control session from your project directory:
claude remote-control
The terminal displays a session URL and a QR code. You have three ways to connect from another device:
- Scan the QR code with the Claude iOS or Android app (press spacebar to toggle the QR display)
- Open the session URL in any browser at
claude.ai/code - Find the session by name in the Claude app or web interface -- active Remote Control sessions show a computer icon with a green status dot
If you are already mid-session and want to enable Remote Control without starting over, use the slash command:
/remote-control
Or the shorthand:
/rc
This carries over your entire conversation history. One practical tip from the documentation: use /rename before activating Remote Control to give your session a descriptive name. When you have multiple sessions running across different projects, "Refactoring auth module" is easier to find than "Remote Control session."
To enable Remote Control automatically for every future session, run /config inside Claude Code and set Enable Remote Control for all sessions to true.
Pro Tip: Use the /mobile command inside Claude Code to display a download QR code for the iOS or Android app. Useful for getting the app installed quickly on a new device.
The security model: outbound-only, short-lived credentials
The most common reaction to "control your terminal from your phone" is a security wince. Anthropic designed Remote Control to address this with a specific architectural choice: the local process never opens a listening port.
Here is how the connection works:
- Your local Claude Code process makes an outbound HTTPS request to the Anthropic API and registers the session
- The process then polls the API for incoming messages -- it reaches out, rather than waiting to be reached
- When you connect from your phone, the API server routes messages between your mobile client and your local session over TLS
- The connection uses multiple short-lived credentials, each scoped to a single purpose and expiring independently
Anthropic's documentation states this design "limits the blast radius of any single compromised credential." Because the local machine only makes outbound connections, there is no attack surface from the network side -- no open ports to scan, no services to probe.
If your laptop sleeps or your Wi-Fi drops, the session automatically reconnects when the machine comes back online. There is one timeout to be aware of: if the machine is awake but unable to reach the network for more than approximately 10 minutes, the session times out and the process exits. You will need to run claude remote-control again.
Worth noting: The --dangerously-skip-permissions flag, which bypasses approval prompts in normal Claude Code sessions, is deliberately not supported with Remote Control. Every file edit, test run, and command execution requires your explicit approval from whatever device you are connected on. This is a deliberate security decision -- Anthropic does not want unattended Remote Control sessions making changes without human oversight.
Remote MCP servers: Sentry, Linear, and Figma plug in without local installation
Alongside Remote Control, Anthropic launched support for remote MCP (Model Context Protocol) servers. This is a separate feature but directly complementary.
Previously, connecting a third-party tool to Claude Code meant running a local MCP server -- installing packages, managing processes, handling authentication yourself. Remote MCP servers flip this: the tool provider hosts the server, and you connect with a single command. Native OAuth handles authentication, so there are no API keys to manage or credentials to store.
Three launch partners are live:
Sentry (error monitoring):
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp
Gives Claude direct access to your Sentry issues, errors, projects, and Seer analysis. Claude can investigate error patterns across your application stack and suggest fixes based on error context.
Linear (project management):
claude mcp add --transport sse linear https://mcp.linear.app/sse
Create issues from code discussions, search tickets, update statuses, and add comments -- all without leaving your terminal. Tom Moor, Linear's Head of Engineering, described the integration: "Engineers can now stay in flow while maintaining full visibility into their project's context."
Figma (design):
claude mcp add --transport http figma-remote-mcp https://mcp.figma.com/mcp
Pull design context from Figma designs, FigJam boards, and Make files directly into your coding session.
MCP servers can be installed at three scopes: local (private to you in the current project, the default), project (shared via .mcp.json for team collaboration), or user (available across all your projects). Use /mcp inside Claude Code to manage servers and authenticate via OAuth.
Unlike Remote Control, remote MCP servers are available on Pro, Max, Team, and Enterprise plans.
The OpenClaw comparison: different tools, different bets
Remote Control did not ship in a vacuum. It arrived during the same week that OpenClaw -- the open-source personal AI agent created by Peter Steinberger -- continued to dominate developer conversations. OpenClaw has accumulated over 220,000 GitHub stars (the fastest-growing repository in GitHub history) and operates through messaging apps like WhatsApp, Telegram, and Discord with 700+ community-built skills.
The tools are solving fundamentally different problems.
Claude Code Remote Control is a scoped, developer-focused tool: it lets you control a local coding agent from your phone with enterprise-grade security. It runs one model (Claude), on one platform (Anthropic's), within one security perimeter.
OpenClaw is a general-purpose life agent: it manages files, browsers, calendars, email, smart homes, and code across any AI model (Claude, GPT, DeepSeek, or local models). It bets on breadth, extensibility, and open-source community innovation.
The security stories could not be more different. Claude Code's controlled ecosystem means Anthropic manages the entire trust chain -- outbound-only connections, short-lived credentials, mandatory permission approvals. OpenClaw's open skill marketplace has faced serious security challenges. Cisco's AI security research team found that ClawHub's top-ranked skill was silently exfiltrating data to attacker-controlled servers. Koi Security identified 341 malicious skills in the ClawHub ecosystem, including the ClawHavoc campaign that distributed infostealer packages.
| Dimension | Claude Code Remote Control | OpenClaw |
|---|---|---|
| Scope | Coding-specific, cross-device | General-purpose life agent |
| Model support | Claude only | Model-agnostic |
| Cost | $100-$200/month (Max plan) | Free and open-source |
| Security | Enterprise-grade, controlled | Community-driven, documented vulnerabilities |
| Best for | Professional developers mid-task | Tech enthusiasts wanting a personal AI agent |
Steinberger himself joined OpenAI on February 14, 2026, to lead its personal AI agent efforts -- signaling that the "always-on agent accessible from any device" paradigm is the next competitive frontier for every major AI company.
Known limitations: still a research preview
Anthropic labels Remote Control as a "research preview," and hands-on testing confirms the label is accurate. Simon Willison, the Django co-creator and AI tools writer, documented several issues during his testing:
- First-launch authentication bug: Remote Control initially returned "not enabled for your account" even for Max subscribers. The fix is to log out (
/logout) and log back in (/login). This is a known bug tracked on GitHub. - One session at a time: Each Claude Code instance supports exactly one remote connection. You can run multiple Claude Code instances (one per project), but each gets only one remote session.
- Terminal must stay open: Remote Control runs as a local process. Close the terminal or stop the
claudeprocess and the session ends. Laptop sleep is fine -- it reconnects automatically -- but process termination is final. - API stability: Willison hit API 500 errors during his session, and restarted sessions produced "mysterious API errors rather than neatly explaining that the session has terminated."
- MCP servers cannot be added from mobile: You must configure MCP integrations from the terminal first. The mobile interface is for monitoring and approving actions, not for configuration.
- No Team or Enterprise plans yet: Only individual Max subscribers have access during the research preview. Pro plan support is listed as "coming soon" with no specific date.
Worth noting: The feature requires a Max subscription at $100 or $200/month. For individual developers on the $20 Pro plan, this is currently out of reach. Anthropic has confirmed Pro access is coming but has not committed to a timeline.
What this changes for developer workflows
Remote Control solves a specific pain point that every developer who uses agentic coding tools has experienced: the tethering problem. When Claude Code is mid-task -- running a migration, debugging a test suite, refactoring across dozens of files -- you are chained to your laptop. Get up and the context is frozen. Come back and you have lost the thread.
Remote Control breaks that chain. Three practical scenarios illustrate the shift:
Long-running refactors: Start a CSS-to-Tailwind migration across 40+ files. Enable Remote Control. Walk to get coffee. Approve or reject changes from your phone as Claude processes each file. The migration keeps moving while you keep moving.
Build monitoring: Kick off a failing test investigation. Claude reads logs, identifies the issue, and proposes a fix. You review and approve the fix from your phone while you are in a meeting. The build goes green without you returning to your desk.
Multi-project management: Run three separate Claude Code instances -- one for a bug fix, one for test coverage, one for a dependency upgrade. Each has its own Remote Control session. Switch between them on your phone like switching chat threads.
The underlying shift is that agentic coding tools are moving from "session-based interactions you sit in front of" to "persistent agents you check in on." Remote Control is Anthropic's first step in that direction. The fact that Anthropic simultaneously announced scheduling features for Cowork (Claude's general agent for knowledge work) suggests this is a deliberate company-wide strategy: AI agents that run in the background and surface decisions to humans wherever they are.
The Bottom Line
Claude Code Remote Control is a narrowly scoped, well-engineered feature that solves a real problem. It is not a mobile IDE. It is not a cloud development environment. It is a secure bridge between a local terminal process and whatever device you have in your pocket.
The architecture is sound -- outbound-only connections, short-lived credentials, no open ports, mandatory permission approvals. The execution is rough around the edges -- authentication bugs, API 500 errors, and a $100/month minimum price tag mean this is genuinely a research preview, not a finished product.
But the direction is clear. The era of AI coding agents as stationary, sit-down tools is ending. Remote Control is the first indication of where Anthropic thinks development is going: agents that run locally, execute autonomously within defined guardrails, and surface decisions to the developer on whatever device is closest.
For developers already embedded in the Claude Code ecosystem, this is worth trying today. For everyone else, this is worth watching closely. The most interesting question is not whether you can code from your phone -- it is what happens when you no longer need to be at your desk for your AI to keep working.
Sources
- Anthropic Blog: Remote MCP Support in Claude Code (Feb 25, 2026)
- Claude Code Documentation: Remote Control (Feb 2026)
- Claude Code Documentation: Security (Feb 2026)
- Claude Code Documentation: MCP (Feb 2026)
- VentureBeat: Anthropic Just Released a Mobile Version of Claude Code Called Remote Control (Feb 25, 2026)
- Simon Willison: Claude Code Remote Control (Feb 25, 2026)
- DevOps.com: Claude Code Remote Control Keeps Your Agent Local and Puts It in Your Pocket (Feb 25, 2026)
- The New Stack: Anthropic's Claude Code Gets Support for Remote MCP Servers (Feb 25, 2026)
- Help Net Security: Anthropic Remote Control Claude Code Feature (Feb 25, 2026)
- Unite.AI: OpenClaw vs Claude Code Remote Control Agents (Feb 2026)
- Cisco Blog: Personal AI Agents Like OpenClaw Are a Security Nightmare (Feb 2026)