AMC (Agent MCP Commander)
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@AMC (Agent MCP Commander)show me the files in ~/projects on my laptop"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
AMC — Agent MCP Commander
Let your AI apps safely work on your own computers.
AMC is a small, self-hosted relay that gives AI apps — Claude Code, Codex, Claude.ai, ChatGPT or any MCP client — a single MCP endpoint for all of your machines. Each machine runs a tiny device agent that connects out to the relay, so you never open a port on your laptop, desktop or server. Every call is checked against the key the AI app was given (which machines, which folders, read/write/run) and written to an audit log.
Claude Code ─┐ ┌──> laptop (device agent)
Codex ───────┤ MCP (HTTPS + key/OAuth) │
Claude.ai ───┼──────────────────────> AMC relay ─┼──> desktop (device agent)
ChatGPT ─────┤ (you host) │
any MCP app ─┘ └──> server (device agent)
^
devices connect OUTBOUND ──┘ (WebSocket; no inbound ports on devices)Install in one line
macOS / Linux
curl -fsSL https://raw.githubusercontent.com/inflow-digital/amc/main/installers/install.sh | shWindows (PowerShell, no admin needed)
irm https://raw.githubusercontent.com/inflow-digital/amc/main/installers/install.ps1 | iexThe installer gets uv if you don't have it (it brings its
own Python), installs the amc command and runs amc up, which:
creates the relay on this machine (listening on
127.0.0.1:8790by default);starts it as a per-user background service (systemd
--user, a launchd LaunchAgent, or a Scheduled Task at logon on Windows);pairs this machine to it as a device, allowed to use your home folder;
connects Claude Code and Codex automatically if they are installed;
runs
amc doctorto prove everything works.
Running the installer (or amc up) again is safe: it upgrades and repairs.
Related MCP server: remote-desktop-commander
Add another machine
On the machine running the relay:
amc relay pair --name my-desktopIt prints a one-time code (valid 15 minutes) and the exact command to paste on the new machine, for example:
curl -fsSL https://raw.githubusercontent.com/inflow-digital/amc/main/installers/install.sh \
| sh -s -- --pair https://relay.example.com ABCD-2345The new machine must be able to reach the relay. If the relay only listens on
127.0.0.1, either give it a private-network address (amc relay init --bind 0.0.0.0 --force,
e.g. on a VPN such as Tailscale) or an HTTPS URL (see below).
Connect your AI
App | Command | What happens |
Claude Code |
| automatic ( |
Codex |
| automatic (writes |
Claude.ai / Claude Desktop |
| prints the URL to add as a custom connector and a key to paste on the AMC login page |
ChatGPT |
| same, for a ChatGPT connector (developer mode) |
Any MCP client |
| prints the Streamable HTTP URL and the |
Keys are shown once. Running amc connect ... again rotates that app's key.
Claude.ai and ChatGPT need HTTPS, because they connect from the internet. Two easy ways that need no router changes:
cloudflared tunnel --url http://127.0.0.1:8790 # prints https://<random>.trycloudflare.com
tailscale funnel 8790 # https://<machine>.<tailnet>.ts.netThen tell AMC its public address and connect again:
amc relay url --public-url https://relay.example.com
amc connect claude-web # or: amc connect chatgptA quick tunnel's URL changes every time it restarts; use a named tunnel or Tailscale Funnel for something permanent.
Let your AI install it
Claude Code — add the plugin, then just ask "install AMC":
/plugin marketplace add inflow-digital/amc
/plugin install amc@amcCodex (or any agent that reads AGENTS.md) — point it at
skills/install-amc/AGENTS.md and ask it to follow it.
Security model
AMC is built so that the relay decides and the devices do the work — and each side checks again.
Keys and access levels. Every AI app gets its own agent key. A key has an access level and can be limited to some devices (
--devices) and folders (--paths):Level
Allows
readread files, list folders, search, list processes, system status
standardread+ write, edit, move files and create foldersfullstandard+ run commands and stop processesamc key add NAME --access read --devices laptop --paths /home/alice/projectscreates a narrow one (useamc device listfor device ids;--pathstakes absolute paths).amc key remove NAMErevokes it immediately, including OAuth sessions it approved.What
fullreally means. Running commands (start_process) runs a shell as your user account. Folder limits cannot contain a shell: afullkey can do anything you can do on that machine. Givefullonly to apps you trust with your account; usestandardorreadotherwise. AMC is not a sandbox.Device roots. Each device only touches files inside its allowed folders (
--root, default: your home folder). The device resolves~, relative paths and symlinks before checking, so..or a symlink cannot escape.Outbound-only devices. Devices open one WebSocket to the relay. Nothing listens on them, so there is nothing to port-scan or expose.
Hashed secrets. Agent keys, device tokens, pairing codes and OAuth tokens are stored only as SHA-256 hashes (files are readable only by you). Pairing codes are single use and expire; failed attempts are rate-limited.
Audit log. Every allowed or denied call is written to
audit.jsonlwith who, which device, which tool and the decision — never arguments, file contents, results or secrets. An app can read its own events with theremote_audittool.Environment isolation. Commands and stdio MCP servers on a device get only an allowlist of ordinary environment variables (
PATH,HOME, locale, ...). The device token and anyAMC_*variables are never passed on.Use HTTPS beyond a private network. Keys travel as bearer tokens. Plain
http://is fine on127.0.0.1or inside a VPN such as Tailscale; anywhere else, put the relay behind HTTPS (a tunnel or a reverse proxy).
Details: docs/security.md. Report vulnerabilities privately — see SECURITY.md.
Troubleshooting
amc doctor # checks everything; every failure prints the command that fixes it
amc relay status # devices online/offline, keys
amc host status # this machine's pairing
amc service status relay # or: host — running / stopped / not installedService logs are in the AMC config folder under logs/ (relay.log, host.log):
~/.config/amc on Linux, ~/Library/Application Support/amc on macOS, %APPDATA%\amc
on Windows.
Uninstall
amc service uninstall host
amc service uninstall relay
uv tool uninstall amc-commanderAlso delete the config folder above to forget keys and device identities.
Documentation
docs/cli.md — every
amccommanddocs/architecture.md — components, protocol, tool catalog, auth
docs/security.md — threat model and known limitations
CONTRIBUTING.md · THIRD_PARTY.md · LICENSE (MIT)
This server cannot be deployed
Maintenance
Related MCP Connectors
Connect any AI agent to 1,000+ apps and 27,000+ actions through one remote MCP server (OAuth).
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceProvides a secure MCP gateway for AI agents to access APIs without exposing raw credentials, with scoped access, audit logging, and OAuth support.MIT
- FlicenseNot gradedqualityCmaintenanceEnables secure remote access to your computer's filesystem and terminal through MCP, allowing AI assistants to manage files, run commands, and automate tasks from anywhere via a hosted relay.105-

Lanes Linkofficial
AlicenseNot gradedqualityBmaintenanceOne secure, self-hosted MCP endpoint connecting AI agents to your real accounts (mail, calendar, files, notes), with per-profile separation, explicit permission enforcement, and an append-only audit log.146 npm35Apache 2.0- AlicenseNot gradedqualityAmaintenanceEnables self-hosted AI gateway and agent control plane with governed MCP tools, virtual-key budgets, caching, and audit, supporting OpenAI, Anthropic, Gemini, MCP, and A2A protocols. It provides deterministic prompt enhancement and governed execution with a zero-credential first run.8Apache 2.0