SkillAgentSearch skills...

control

🎛️ Run commands, AI agents, and workflows across your Windows, Linux, and macOS machines. Peer-to-peer execution and file transfers with CLI, MCP, and HTTP API access.

Install / Use

claude mcp add koltyakov -- npx -y github:koltyakov/control

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

81/100

Category

Automation

Supported Platforms

Claude Code
Claude Desktop

Our assessment of control

control scores 81/100 on our quality scale, 2406th of 2,867 Automation skills we index.

Its MCP Server is 10 KB long, well organised into 13 sections with 10 code examples: a thorough specification that gives an agent plenty to work with.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
29/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so control is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

control compared with similar skills

All 4 of these similar skills score higher than control; compare them before choosing.

SkillScoreStarsUpdatedFormat
control (this skill)by koltyakov813todayMCP Server
Agent-Reachby Panniantong100108.1k2d agoCLAUDE.md
headroomby headroomlabs-ai10078.7ktodayCLAUDE.md
CowAgentby zhayujie10047.8k1d agoCLAUDE.md
Scraplingby D4Vinci10093.4ktodayMCP Server

Frequently asked questions

How do I install control?
Run claude mcp add koltyakov -- npx -y github:koltyakov/control. The install tabs above show the steps for each supported agent.
Which AI agents does control work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is control safe to use?
It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is control still maintained?
The repository was last updated today, so control is actively maintained.

Control

<p align="center"> <img src="./assets/logo.png" alt="Control logo" width="220" /> </p>

A Go peer execution network for Windows, Linux, and macOS. Connect machines to a gateway and address them by name through CLI or MCP. A local HTTP API handles node-local operations.

Orchestrators coordinate work; worker nodes execute it and exchange files directly. These are operation roles, not machine types. The gateway handles enrollment, discovery, connections, and fleet administration. AI agents are optional.

What it does

  • Private user fleets with persistent machine identities and names.
  • Direct WebRTC connections with WebSocket relay fallback and mutually authenticated TLS on both paths.
  • Durable tasks with idempotent submission, cancellation, streamed logs, and restart reconciliation.
  • Commands, scripts, AI CLIs, MCP servers, and opt-in native desktop automation with worker-local secret references.
  • SHA-256 artifacts, resumable downloads, and direct worker-to-worker delivery.
  • Forward/reverse TCP tunnels, private-network HTTP requests, and explicit clipboard pastes.
  • Wake-on-LAN for offline machines through a same-LAN node or an explicitly selected local client.
  • Leases, bounded parallel workflows, terminal and web dashboards, and idle-only managed updates.

Account-authenticated clients initiate execution. Workers need instruction-bound delegation to request peer work; fleet membership permits discovery only. See authority and delegation.

For bounded task listings use control task list NODE --limit 100, then pass the returned nextCursor with --cursor. control task reconcile NODE ID inspects an interrupted workflow's child tasks without replaying them. control diagnostics NODE reports storage, cleanup, MCP and transport counts. See task recovery and limits, artifact budgets, and signed release verification.

Architecture

Architecture

See architecture for roles and connections.

Install the gateway on a Linux VPS

Follow the Linux VPS guide for Ubuntu/Debian setup with systemd, Caddy HTTPS, and user accounts.

Already running expose? Use the expose gateway guide instead of Caddy.

Point a domain at the VPS, allow TCP 80/443, and keep the gateway on 127.0.0.1:7330 behind the proxy. Create accounts with control users create NAME. Give users their own account keys, never the gateway superuser key.

The gateway URL also serves the web console. Set a separate browser password with control console password; operators sign in as @operator. Operators can provision an existing user's browser login with control console password NAME. CLI account keys are not browser passwords.

Install the orchestrator CLI

Linux or macOS:

curl -fsSL https://raw.githubusercontent.com/koltyakov/control/main/scripts/install.sh | sh

Windows, from PowerShell, CMD, or Bash:

powershell.exe -NoProfile -c "irm https://raw.githubusercontent.com/koltyakov/control/main/scripts/install.ps1|iex"

Installers require published GitHub releases. On Windows, open a new terminal to reload PATH.

Log in with the account key supplied by your gateway operator:

control login --gateway https://control.example.com
control install-mcp opencode
control install-skill opencode
control machines
control dashboard

Login saves the validated URL and key. The release installer already installs the OpenCode skill; MCP and other client integrations are optional. See other clients. Remote work needs no local node:

control exec worker -- hostname
control call worker node.describe

To also run this machine as a worker:

control setup --name main --client opencode

Setup reuses your login and installs node startup, MCP, and the skill. On Windows, it requires Administrator PowerShell by default; --service user selects startup at login under your normal account.

Install workers on Windows, macOS, and Linux

control machines add win-worker --platform windows --service user
control machines add mac-worker --platform macos --service user
control machines add linux-worker --platform linux --service system
control machines add workstation --platform linux --service hybrid

Run each printed command on its target under the intended user. Windows uses PowerShell and requests UAC for firewall setup. macOS uses Bash; Linux system mode needs sudo and systemd. Installers detect amd64/arm64, verify the binary, and enroll and start the node without an account key. Verify with control machines and control call win-worker node.describe.

Use auto or omit the name for the target's hostname. Installation URLs use an 8-character, single-use ticket and expire after 15 minutes by default. Nodes generate separate full-length security credentials. Press a in the dashboard to add a machine. See platform requirements.

Use --force when adding or renaming to retract a conflicting pending invitation, never an enrolled identity.

User context has the installing user's files and credentials but may stop at logout. System context starts at boot as root on Linux/macOS or LocalService on Windows. See startup contexts.

Hybrid installs isolated User and System profiles. Address workstation for User work or explicitly choose workstation-system for System work. The System companion runs as root or Windows LocalSystem and requires exact instruction grants. It is never selected automatically or used as a fallback. See Hybrid installation.

Run and inspect work

control exec worker --detach --id job-001 --timeout 4h -- COMMAND ARG...
control task logs worker job-001 --follow
control task wait worker job-001
control task cancel worker job-001

Stopping a wait or log follower does not cancel work. Node restarts do not replay interrupted tasks. Use a new task ID for new work. Finished records, logs, and workspaces are retained for seven days and up to 10,000 tasks by default; pruning and reconciliation describe the remaining tombstones.

control system worker
control system worker --refresh
control speedtest worker --size 64 --json
control artifact export worker input.mp4
control artifact get worker ARTIFACT_ID ./output.mp4

Dashboard polls read cached resource samples; connection tests run on request. Artifact downloads resume and verify SHA-256. See the protocol reference for calls, grants, and workflows.

For repeatable work, keep shared or machine-specific action instructions in runbooks. Client node descriptions combine project/global runbooks with authorized worker stores. Use control runbooks, add --node NAME for worker discovery, or use MCP control_runbooks. Discovery never executes instructions or retrieves secret values.

Install MCP and skills

Install either or both on the orchestrator. MCP exposes tools; the skill teaches CLI usage. Both install globally by default.

control install-mcp opencode
control install-skill opencode

Replace opencode with codex, claude, cursor, copilot, windsurf, antigravity, or agents for other clients. Add --project . for project-local installation. Restart the AI client afterward.

The agents target uses ~/.agents/mcp.json and ~/.agents/skills/control/SKILL.md, or project-local equivalents. Manual MCP import may be needed. See configuration paths.

For other MCP clients, configure a stdio server running /absolute/path/to/control mcp. It uses your saved login, without a local node. Explicit --api or CONTROL_API selects local-node authority, not independent peer execution.

Build and test

Requires Go 1.27.3 or newer. No CGO dependency.

make build
make fmt-check
make lint
make check
make test-compose

Compose tests run a gateway, three nodes, and an FFmpeg workflow over WebRTC and relay. Docker supplies dependencies. See make help and testing for other targets and benchmarks.

Updates

control update updates the CLI from GitHub Releases. Managed node restarts wait for idle reservations. With saved superuser authorization, make update pushes a development bundle to the gateway and nodes; make update-status tracks it. See managed updates.

Managed rollouts switch the gateway first, then nodes. Offline nodes catch up on startup without blocking the available fleet indefinitely. Nodes prefer an exact signed GitHub release and use the gateway's retained bundle when GitHub is unavailable or its assets differ.

Documentation

Limits and trust

One gateway, no clustering, public signup, billing, or execution sandbox. Providers have their node's OS permissions. Keep account keys and gateway state inaccessible to workers, and the local API on loopback. Trust the gateway for naming, membership, and executable distribution. See security boundaries.

Licence

MIT

Related Skills

View on GitHub
GitHub Stars3
CategoryAutomation
Updated1h ago
Forks1

Languages

Go

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low