husk
Connects the server to Discord as a bot, letting users run commands, inspect tasks, and receive results through server or direct message chats.
Lets the server provision and manage disposable Linux machines using Docker containers, providing the agent with a shell, filesystem, network, and ports within an isolated container.
Allows the server to use Google AI models (e.g., Gemini via Google AI Studio) as the agent's model backend for reasoning and planning.
Allows the server to use local Ollama models (e.g., Gemma, Llama, Qwen) as the reasoning backend, enabling free, private, and local operation without an API key.
Allows the server to use OpenAI models for agent reasoning and task planning, expanding the model flexibility beyond local options.
Lets the server provision and manage disposable Linux machines using Podman containers, offering a rootless and daemonless alternative to Docker for isolated agent environments.
Enables the server to run as a Slack bot, allowing users to interact with the disposable computer and task execution directly in Slack channels and DMs.
Enables the server to operate as a Telegram bot, allowing users to send commands and view task outputs within Telegram chats.
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., "@huskspin up a disposable Linux box and show me the OS details"
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.
Husk gives AI agents disposable Linux computers — a shell, a filesystem, network, ports — that they can drive and you can throw away. It also turns any AI chat transcript into a reusable bot you can read, edit, and run as a service.
Interactive architecture diagram — open locally, no server needed. Pan, zoom, dark mode, PNG/SVG export.
Free. Runs on your hardware. No account required.
Quick start
npx @husk/cli doctor # what's available on this machine
npx @husk/cli up dev # spin up a Linux computer
npx @husk/cli exec dev -- 'uname -sr && python3 -V'
npx @husk/cli rm dev # tear it downNo Docker? It still works — see what each provider gives you.
Give Claude Code a computer
claude mcp add husk -- npx -y @husk/mcpOne command. Claude Code gets shell, read_file, write_file, list_dir and expose_port against a real Linux machine — sandboxed away from your repo, filesystem persisting across the conversation. Works with Cursor, Zed, or anything that speaks MCP.
Turn a chat into a bot
husk import # finds Claude Code / ChatGPT transcripts
husk distill 3 --out triage.yaml # conversation → agent spec
husk run triage.yaml "check the build"What comes out is a YAML file you can read and version:
name: triage
model: sonnet
persona: |
You triage CI failures for a TypeScript monorepo.
Always read the failing job log before guessing.
tools: [computer, files]
computer:
flavor: node
network: { mode: egress, allow: ['*.github.com'] }
limits: { maxSteps: 24, maxCostUsd: 0.25 }Related MCP server: linux-harness-mcp
Features
5 computer providers — Docker, Podman, local (WSL2/POSIX), SSH, Fly
10+ model providers — Ollama, Anthropic, OpenAI, Google, Groq, DeepSeek, Mistral, Together, LM Studio, and more
MCP server — expose a computer as a tool for any MCP-compatible client
Chat-to-bot pipeline — import transcripts, distill to
husk.yaml, run or serveAdapters — Discord, Slack, Telegram, webhook out of the box
Browser automation — Chromium lifecycle, CDP, screenshots, click/type/scroll
Zero config path — no Docker, no API key, no account: everything still runs
No telemetry — not "off by default" — absent. No analytics, no crash reporter, no phone-home
No native modules —
npm installis clean on Windows without a C++ toolchain
Models
Bring whatever you have. Aliases resolve across providers so the same husk runs on Opus or on a local Gemma.
Free, local | Ollama ( |
Free tier | Groq, Google AI Studio, OpenRouter |
Paid | Anthropic ( |
Isolation
This is the part most tools are vague about, so here it is plainly.
Provider | Isolation | Cost | Notes |
| Kernel namespaces, cgroups, seccomp, read-only root | Free | Default when daemon is up |
| Kernel, rootless | Free | Linux without Docker |
| Guardrails only | Free | Real Linux via private mount namespace |
| Guardrails only | Free | Your shell, jailed to a workspace |
| Whatever the remote provides | Free if you own it | Oracle Free Tier, a Pi, a VPS |
| microVM | Metered | Bursty parallel work |
The
localprovider is not a sandbox. It stops accidents, not adversaries.husk doctorreportsisolated: falsefor it. See SECURITY.md for the full model.
Install
npm i -g @husk/cli # or: npx @husk/cliNode 20.10+. No native modules. Docker optional. API key optional.
Packages
packages/
core contracts, husk.yaml schema, primitives
runtime computer providers: docker, podman, local, ssh, fly
models one surface over ten model providers
sessions transcript importers + the distiller
browser Chromium lifecycle and CDP automation
agent tool-calling loop and built-in tools
adapters Discord, Slack, Telegram, webhook
mcp MCP server (stdio + streamable HTTP)
server control-plane API + bot host
sdk typed client for the control plane
cli the husk command
apps/
console dashboard: live terminal, files, browser
docs documentation site (Next.js + MDX)
web marketing site (Next.js + Three.js)Documentation
Architecture — why it is shaped this way
API reference — the control-plane HTTP contract
Build contract — conventions every package obeys
Security model — what is isolated and what is not
Examples — six ready-to-run
husk.yamlrecipesDiagrams — 16 interactive architecture, dataflow and design diagrams (open the
.htmlfiles locally)
Development
git clone https://github.com/Hotragn/husk.git && cd husk
npm install
npm run build
npm testSee CONTRIBUTING.md for conventions, test philosophy, and how to add a provider or model.
Contributing
Contributions are welcome. Please read CONTRIBUTING.md before opening a PR — it covers the build contract, test requirements, and the values that show up in code review.
License
This server cannot be deployed
Maintenance
Related MCP Connectors
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to remotely read/write files and execute commands on Linux servers via MCP protocol.1 npmMIT
- AlicenseNot gradedqualityCmaintenanceExposes a Linux host with shell, Python, and filesystem tools as MCP tools over HTTPS, enabling AI agents to execute commands and manage files remotely.14 npmMIT
- FlicenseNot gradedqualityCmaintenanceEnables AI agents to operate a persistent Linux sandbox in the cloud, running commands, managing files, using Git, and publishing artifacts through a Streamable HTTP MCP endpoint.-
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to operate a real, private, local Docker-based computer with durable files, terminal access, web research, and a persistent browser or desktop. It supports multiple agent clients via MCP or OpenAPI, with live viewing and human takeover.5Apache 2.0