Give your agent a token.
Not an SSH key.

rcmd runs commands on your servers without SSH, open ports or a VPN. Now your AI agent can too — scoped to the machines you name, expiring when you say, with every command it runs written down.

Still a first-class CLI for humans: exec, file copy faster than scp, tunnels that replace ssh -L, multi-target fan-out.

Start with an agent View on GitHub

Your agent, on a short leash

An agent that can run commands on production needs limits an SSH key cannot express. One command mints a token that has them.

$ rcmd agent-token --targets web1 --expires 4h --label claude

Token: tok_operator_9f2c…
Targets: web1  Expires: in 4h0m
MCP config, ready to paste — and the revoke command, right there.

Scoped to named machines

An agent token must name the hosts it may touch — there is no accidental fleet-wide token. Try it on anything else and the relay refuses, and says why, so the agent stops retrying instead of hammering.

Expires on its own

Four hours, a day, whatever the task needs. A token that outlives the job is a liability, and an agent is not around later to notice it still works.

Read-only is a real option

Give a viewer token and the agent can look — list machines, check health — and cannot execute a thing. Investigation without exposure.

Everything is written down

Every command, file transfer, tunnel and scheduled job lands in one audit trail: who, where, when, and the exit code it actually returned. Yours, your teammates', your agents'.

A kill switch you can reach

rcmd team revoke takes effect immediately. rcmd team list shows what can currently touch your fleet and how long each has left.

Rate limited by default

A retry loop with no backoff is the normal failure mode of an autonomous agent. The relay throttles per token and tells the caller how long to wait, so one noisy agent cannot take a host down or starve a teammate.

Speaks MCP — the 2026-07-28 revision, and the older handshake for clients that have not migrated. Works with Claude Code, Claude Desktop, and anything else that speaks MCP.

What you get

Fast file transfer

rcmd cp streams files through the relay with 2 MiB chunks, adaptive gzip, and parallel streams — benchmarked faster than scp over high-RTT links. Directories auto-tarred. No base64, no buffering.

Port forwarding

rcmd tunnel replaces ssh -L. Expose a remote port locally through the relay — reach that database behind NAT without a VPN or a bastion.

Multi-target fan-out

Run one command across every server at once. rcmd exec --targets web1,web2,db1 --cmd "apt update" — results back as JSON, one per target.

MCP server built in

rcmd mcp exposes exec, copy, health and target discovery as MCP tools, scoped by the token you hand over. Speaks the 2026-07-28 revision and the older handshake, so it works with clients that have not migrated yet. How it works.

AI-agent native

Every command returns structured JSON — stdout, stderr, exit code, duration. Deterministic timeouts. No parsing, no guessing, no wasted tokens.

Scheduled commands

Cron without cron. rcmd cron add runs commands on a schedule from the relay — "restart nginx every day at 2am" without touching crontab on each box. Email alerts on failure.

Team access

Scoped sub-tokens with admin, operator, and viewer roles. Email invites, audit logs. Share server access without sharing your token or your SSH keys.

How it works

1. Install

One curl command installs the daemon on any Linux server. No dependencies, no runtime, no config.

2. Connect

The daemon connects out to the relay. Zero inbound ports. Works behind NAT, CGNAT, firewalls, corporate proxies.

3. Execute

Run commands, copy files, open tunnels from your terminal. The relay routes everything. JSON in, JSON out.

Install in one command

curl -sSL https://rcmd.intrane.fr/install.sh | sh

Installs the rcmd CLI — then run rcmd signup --free --email you@company.com (no credit card) or rcmd signup --email you@company.com (Pro)

Pricing

Free

0/mo
Hosted relay. No credit card. Get started →
  • 3 connected servers
  • Hosted relay at rcmd.intrane.fr
  • Command execution (exec, cp, multi-target)
  • Streaming output, JSON output
  • Automatic TLS (Let's Encrypt)

Pro

5/mo
For teams and growing fleets. Everything in Free, plus:
  • Unlimited connected servers
  • Scheduled commands (cron)
  • Port forwarding (tunnels)
  • Team access — scoped sub-tokens, roles, audit logs
  • Email alerts on failures
  • Webhooks — HMAC-signed event callbacks

Self-hosted

0/mo
Run your own relay. Full control. Setup guide →
  • Unlimited targets
  • All features unlocked
  • Run on any VPS, your infrastructure
  • Open source (MIT)

FAQ

How is this different from SSH?

SSH requires open ports, key management, and VPN for NAT traversal. rcmd's daemon connects out to the relay — no inbound ports, no keys, no VPN. It works from anywhere.

What's the difference between Free, Pro, and Self-hosted?

Free gives you 3 servers on our hosted relay — no credit card, no setup. Pro removes the server limit and adds cron, tunnels, team access, alerts, and webhooks. Self-hosted is the open-source binary — you run the relay on your own VPS with everything unlocked.

Is it secure?

Each target has a unique token. All traffic is over WebSocket with optional TLS. The relay never stores commands or output — it just routes traffic. You can also self-host the relay for full control.

Can AI agents use it?

Yes. rcmd was designed for AI agents. Every command returns structured JSON (stdout, stderr, exit code, duration). No parsing, no guessing, no wasted tokens.

How do I get support?

Open an issue on GitHub or email arancibiajav@gmail.com.

Can I cancel anytime?

Yes. No contracts, no lock-in. Your daemon keeps working with the free version if you cancel Pro.

Would you use a fleet dashboard?

Every machine on one page — health, history, a shell in the browser. It is a concept, not a product: whether it gets built depends on whether people want it. Two minutes to look, one click to say so.

See the concept