Claude Code Goes Remote: Control Your Development Session From Anywhere
On February 25, 2026, Simon Willison published detailed coverage of a significant new Claude Code feature: the ability to run a persistent local development session on your computer and control it from anywhere—via the web, iOS app, or native desktop client. This fundamentally changes the developer experience, moving from "I must be at my development machine" to "I can guide my agents from anywhere."
The Architecture: Local Session, Cloud Control
The setup is elegant: you start a Claude Code session on your local machine (which has access to your repositories, databases, and dev environment). That session runs persistently. You then control it via Claude.ai web, iOS, or desktop, as if the agent were running in the cloud.
This is technically sophisticated. The local session needs:
- Persistent background process that stays running
- Secure tunnel to cloud infrastructure (for authentication and command relay)
- State synchronization (local execution results flow back to cloud interface)
- Low-latency command dispatch (your web requests trigger local actions in real time)
Anthropic solved these challenges. The result is transparent: you interact with Claude Code via web/mobile as normal, but commands execute against your local environment. No SSH tunnels, no port forwarding, no complicated networking.
Cross-Platform Control
You can start a session on your laptop, then:
- Monitor from your phone: iOS app shows session status, recent outputs, and lets you send follow-up commands
- Review from a browser: Claude.ai web interface displays full session transcript and lets you continue conversation
- Work from desktop: Native desktop apps on Mac and Windows provide richer UI than web
This flexibility is powerful. If your agent is running a long build or test suite, you can monitor progress from your phone while working on something else. If you get a production alert, you can immediately start investigating with Claude Code, even if you're not at your main workstation.
Use Cases That Benefit Most
Certain development workflows become dramatically better:
- Long-running tasks: Start a deployment, deployment test suite, or data migration. Monitor remotely without being tethered to your machine.
- On-call incident response: Get paged at 2 AM about a production bug. Start Claude Code on your laptop, investigate and iterate from your phone in bed.
- Code review collaboration: Share your session with teammates (if supported). Everyone can see the agent's progress and suggest directions via comments.
- Asynchronous development: Start a coding task on your desktop, leave it running, check in from mobile as you move around, continue from your laptop later.
Comparison to OpenClaw's Gateway Architecture
This feature is conceptually similar to OpenClaw's gateway approach. OpenClaw already enables remote access to agents via a central gateway: you call the gateway from anywhere, and it routes requests to your local agents.
The difference:
Claude Code Remote:
- One-to-one mapping: your local machine runs the agent
- Managed authentication and tunneling by Anthropic
- Deep integration with Claude.ai interfaces
- No explicit infrastructure needed beyond Anthropic's cloud
OpenClaw Gateway:
- Many-to-many: your gateway can orchestrate multiple agents and services
- You manage authentication and networking (or use Tailscale for simplicity)
- Can integrate with custom frontends and workflows
- Requires your own infrastructure
Claude Code Remote is simpler and lower operational overhead. OpenClaw Gateway is more flexible and gives you ownership. The right choice depends on your architecture goals.
Security and Authentication
How does Anthropic handle security for remote sessions? You're, in effect, exposing your local environment (terminal access, file system access, repository access) to the cloud.
Anthropic likely implemented:
- Encrypted tunnel: TLS between local session and cloud, so commands and results are not visible in transit
- Authentication: Your Claude.ai account identifies you; Anthropic verifies you own the local session
- Authorization: Only you can control your local session (unless you explicitly share it)
- Audit logging: Commands executed are logged for compliance and debugging
From a security perspective, this is no different than using SSH to your laptop remotely. The sensitive data—your repositories, API keys in .env files, database credentials—is still on your local machine. The tunnel just relays commands and results.
Building Remote Agent Workflows in OpenClaw
If you want similar functionality with OpenClaw, the pattern is:
- Run OpenClaw locally or in a private network
- Deploy a gateway that exposes the agent API securely (using TLS, authentication tokens, etc.)
- Call the gateway from anywhere (web, mobile, desktop)
- The gateway relays requests to local agents and returns results
Anthropic has done the implementation work (tunneling, authentication, cross-platform UI). With OpenClaw, you choose the tradeoff: more operational burden in exchange for full control and flexibility.
Comparison: Remote Access Approaches
If you need remote access to development agents, your options:
Claude Code Remote (native):
- Pros: Simple to set up, Anthropic manages infrastructure, cross-platform support
- Cons: Limited to Claude Code, proprietary, no customization
OpenClaw + Tailscale mesh:
- Pros: Works with any model, full customization, zero-trust networking
- Cons: Requires infrastructure setup and ongoing management
OpenClaw + VPN:
- Pros: Traditional, widely understood approach
- Cons: More operational burden, potentially slower than direct tunneling
Public API (with auth):
- Pros: Works from anywhere with internet, no VPN/mesh setup
- Cons: Data exposed to public internet (though encrypted), larger attack surface
Real-World Scenario: Remote Debugging
It's midnight. You get a Slack alert: production database is running slow. You're not at your desk, but you have your phone. You:
- Open Claude.ai on your phone
- Find your running local Claude Code session
- Ask: "Connect to production database and check slow query log. Identify the top 5 slowest queries."
- Claude Code runs on your laptop (which has SSH access to production), executes the diagnosis, and shows results on your phone
- You see the problem: a missing index on the users table
- You tell Claude: "Add an index on users(email). Test the slow query again."
- Claude executes the remediation and confirms the query is now fast
- You post a quick fix summary to Slack and go back to sleep
This workflow is only possible if your agent can be controlled remotely. Local-only Claude Code would require you to SSH to your laptop, start the agent manually, and iterate through the terminal. Remote control makes incident response efficient.
Privacy Implications
Remote sessions introduce privacy considerations. Your local session traffic (commands and results) flows through Anthropic's infrastructure. What data is logged? How long is it retained? Can Anthropic read your commands?
Anthropic's privacy policy covers this, but it's worth understanding. For many teams, the convenience outweighs the privacy concern. For organizations with strict data residency requirements, remote control of local agents might be restricted by compliance policy.
Scaling Remote Sessions
What if you have many developers, each running local Claude Code sessions? Anthropic needs to:
- Track which local sessions belong to which users
- Route commands to the correct local session
- Handle session timeouts and reconnections
- Scale the relay infrastructure to support thousands of concurrent sessions
For a single developer with one local machine, this is simple. For an enterprise with hundreds of developers, it becomes a systems problem. Anthropic likely has solutions for this, but larger deployments may require additional configuration or licensing.
The Broader Significance
Claude Code Remote represents a shift in how AI development tools are consumed. Rather than agents living in the cloud (sandboxed, constrained), agents now live on developers' machines (with full access to repositories and dev environments) and are controlled remotely.
This is more powerful but also riskier. You're exposing your local environment through the cloud. For teams with strict security practices, this might require additional guardrails (network isolation, command whitelisting, audit logging).
As the feature matures, expect to see rich collaboration features: sharing sessions with teammates, collaborative debugging, pair programming with AI. The foundation is now in place.