mcpdirectory.dev
9,129 servers · updated 23 Sept

Remote Agents MCP server

community Active last commit 30 d
{
  "mcpServers": {
    "47-ronn-remote-agents": {
      "command": "npx",
      "args": [
        "-y",
        "remote-agents@latest"
      ],
      "env": {
        "CLOUDFLARE_API_TOKEN": "<cloudflare_api_token>",
        "REMOTE_AGENTS_TOKEN": "<remote_agents_token>"
      }
    }
  }
}
file .mcp.json (in your project root)install npx -y remote-agentsversion 0.1.40

What it does

An MCP server for controlling fleets of remote machines through AI agents. Agents connect outbound to a relay and can execute commands, manage files, run git operations, and coordinate work across multiple hosts—all over end-to-end encrypted channels. For developers who need to orchestrate remote systems via Claude or other AI interfaces.

  • Run shell commands and execute scripts across a fleet of remote machines and aggregate results by host.
  • Transfer files and sync directories between remote machines using peer-to-peer UDP channels with automatic relay fallback.
  • Schedule distributed MapReduce tasks across tagged subsets of your fleet to partition, process, and reduce data in parallel.

Tools this server exposes

ToolDoes
execExecute shell commands on one or more remote agents, with results aggregated per host.
readRead files from remote machines, across all agents or filtered by tags or OS family.
writeWrite or modify files on remote hosts with optional backup mode and path/command allow-lists.
gitDrive git operations across remote agents—clone, pull, push, commit, or check status on multiple hosts.
send_fileTransfer a single file from one remote agent to another over a direct UDP/QUIC channel.
sync_dirSync directories between remote agents using rsync-like semantics with SHA-256 verification.
map_reducePartition data across the fleet, execute a shell command on each partition, and reduce the outputs with per-partition retry.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.