SkillAgentSearch skills...

localDot

AI assistant with a local Linux desktop, live browser view, and task replay.

Install / Use

claude mcp add Shadowhusky -- npx -y github:Shadowhusky/localDot

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

76/100

Category

Automation

Supported Platforms

Claude Code
Claude Desktop

Our assessment of localDot

localDot scores 76/100 on our quality scale, 2580th of 2,902 Automation skills we index.

Its MCP Server is 8.5 KB long, split into 7 sections with 2 code examples: a thorough specification that gives an agent plenty to work with.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
29/30
Structure
16/20
Description
12/15
Adoption
4/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so localDot 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.

localDot compared with similar skills

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

SkillScoreStarsUpdatedFormat
localDot (this skill)by Shadowhusky7610todayMCP Server
Agent-Reachby Panniantong100101.1k2d agoCLAUDE.md
headroomby headroomlabs-ai10077.6ktodayCLAUDE.md
CowAgentby zhayujie10047.5ktodayCLAUDE.md
Scraplingby D4Vinci10089.3ktodayMCP Server

Frequently asked questions

How do I install localDot?
Run claude mcp add Shadowhusky -- npx -y github:Shadowhusky/localDot. The install tabs above show the steps for each supported agent.
Which AI agents does localDot 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 localDot 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 localDot still maintained?
The repository was last updated today, so localDot is actively maintained.
<p align="center"> <img width="80" src="docs/readme/dot.png" alt="dot, a small white ball with two dark eyes"> </p> <h1 align="center">localDot</h1> <p align="center">An AI with a computer of its own.</p> <p align="center"> <b>English</b> · <a href="README.zh-CN.md">简体中文</a> · <a href="https://localdot.dev">localdot.dev</a> </p>

<a href="https://localdot.dev/film/localdot-film.mp4"><img width="100%" src="docs/readme/film.jpg" alt="The localDot film: six dots, each dressed for its job, under the words A dot for every job"></a>

<p align="center"><a href="https://localdot.dev/film/localdot-film.mp4">▶ Watch the 37-second film</a></p>

dot does real work on a computer of its own, right on yours: a Linux desktop with a browser, a terminal and everyday apps, running in Docker. You watch from your browser or your phone, step in whenever you like, and every task is recorded.

Install

macOS (Apple silicon) or Linux (x64), in Terminal:

curl -fsSL https://downloads.localdot.dev/install.sh | bash

Windows, in PowerShell (alpha, runs through WSL 2):

irm https://downloads.localdot.dev/install.ps1 | iex

The installer checks Docker and walks you through anything missing, then prints a link. Open it, pick a model, and give dot something to do.

localdot open prints a fresh sign-in link, localdot update installs the latest release and keeps your data, and localdot doctor diagnoses a problem. Install guide · Use it from your phone or a server

What dot does

<img width="100%" src="docs/readme/app.png" alt="localDot at work: dot's computer shows Kiyomizu-dera's website in Chromium, dot says what it is doing under the screen, and the chat lists what it found while planning three days in Kyoto">
  • Watch it work. Live, from your laptop or your phone. You see every click and keystroke as it happens.
  • Step in any time. Take control and the mouse is yours. Hand it back and dot picks up where you left off.
  • You decide how often it asks. Custom, Auto or Full access. Even with full access, dot asks before buying anything or deleting things online.
  • Use any model. An API key from 25 providers, your ChatGPT or Claude plan, or a model on your own machine. Switch any time.
  • Rewind anything. Every task is recorded step by step. Scrub back to see exactly what dot did.
  • Share the good part. Trim to the moment, speed it up as much as 16 times, add captions and save it as a video.
  • A dot for every job. Everyday, coder, researcher, finance, video editor or designer. Each brings only the skills its work needs, so it stays focused and spends nothing on the rest.
  • Open source, and yours. The app, dot's computer and every recording stay on your machine, and your keys live in a local vault.
<picture> <source media="(max-width: 600px)" srcset="docs/readme/capabilities-mobile.svg" type="image/svg+xml"> <source media="(max-width: 600px)" srcset="docs/readme/capabilities-mobile.png"> <source srcset="docs/readme/capabilities.svg" type="image/svg+xml"> <img width="100%" src="docs/readme/capabilities.png" alt="Pencil sketch of what dot can do on its own computer: use the screen, a web browser that keeps your sign-ins, a terminal and files; search the web and do deep research with numbered sources; talk with voice; use your apps as tools; run routines on a schedule; keep an editable memory; and ask you before it acts as you."> </picture>

Models

dot needs a model that can read screenshots and use tools. Add one in Settings → Models, then pick it from the message box.

  • API keys: Anthropic, OpenAI, Google Gemini, Azure OpenAI, xAI, Mistral, DeepSeek, Groq, OpenRouter, Together, Fireworks, Cerebras, Perplexity, Cohere, DeepInfra, Moonshot, Alibaba, Z.ai, Zhipu AI, MiniMax, Volcengine Ark and SiliconFlow.
  • Plans: your Claude plan through Claude Code, your ChatGPT plan through Codex, and the MiniMax and Alibaba token and coding plans.
  • Your own: any OpenAI- or Anthropic-compatible server, including Ollama and LM Studio on your machine.

Hosted models see screenshots and the task's context. Model options

Safety

<picture> <source media="(max-width: 600px)" srcset="docs/readme/safety-mobile.svg" type="image/svg+xml"> <source media="(max-width: 600px)" srcset="docs/readme/safety-mobile.png"> <source srcset="docs/readme/safety.svg" type="image/svg+xml"> <img width="100%" src="docs/readme/safety.png" alt="Pencil sketch of six walls between dot and the internet: hard checks, purpose check, approvals, the isolated computer, the network gate and stand-in keys. Searching, reading and clicking pass straight through; sending an email as you asks; posting your notes to a random site asks; poking the router is blocked; smuggling out an API key only gets a stand-in."> </picture>

dot works on its own computer, not yours. Reading, searching, clicking and running code there just happen. How often it asks first is up to you:

  • Custom, the default, asks before dot sends or posts anything as you, pays for something, uploads your files, changes your account's security, or changes or deletes things online.
  • Auto lets a reviewer model approve routine steps. Purchases, sign-ins and online deletions still ask you.
  • Full access gets on with it, and still asks before every purchase and before deleting things online (once per site).

Your LAN, your router and cloud metadata are always blocked. Your real keys stay in keyd, a small local service: dot only ever holds stand-ins, and the real key is added on the way out, for that key's own hosts. Passwords go through a private form straight into the site, so dot never sees them.

Pages, search results and app responses reach the model marked as untrusted. An optional purpose check flags private data heading somewhere unrelated to the task. This is alpha software, so review what dot does with your accounts. How the agent runs · Purpose check

How it works

Your browser talks to the localDot daemon on your machine, which serves the app, runs tasks, streams dot's desktop and records every step. dot's computer is a set of containers: the desktop you watch, a web browser that keeps your sign-ins, the agent's loop, and a gate that is its only way out to the network.

<picture> <source media="(max-width: 600px)" srcset="docs/readme/system-mobile.svg" type="image/svg+xml"> <source media="(max-width: 600px)" srcset="docs/readme/system-mobile.png"> <source srcset="docs/readme/system.svg" type="image/svg+xml"> <img width="100%" src="docs/readme/system.png" alt="Pencil sketch: your browser and phone talk to the localDot daemon on your machine, which runs an isolated computer of four containers: desktop, web browser, brain and gate. The gate leads to keyd, which holds the real keys and adds them on the way out to model providers, the web and your apps. Web search and page reads go through keyd with no sign-ins or keys, and each app that runs as a program gets an app cell of its own."> </picture>

A task goes round a simple loop: act, take a fresh look at the screen, say what happened, and repeat until it's done. When dot needs you, it asks in the chat.

<picture> <source media="(max-width: 600px)" srcset="docs/readme/task-loop-mobile.svg" type="image/svg+xml"> <source media="(max-width: 600px)" srcset="docs/readme/task-loop-mobile.png"> <source srcset="docs/readme/task-loop.svg" type="image/svg+xml"> <img width="100%" src="docs/readme/task-loop.png" alt="Pencil sketch of one task: ask, plan, act, look, narrate, replay. Act, look and narrate repeat until the task is done; when dot needs you, it asks in the chat."> </picture>

| Folder | What's in it | | --- | --- | | apps/daemon | The daemon: API, agent loop, safety checks, tools and recordings (TypeScript on Node) | | apps/web | The app you open in your browser (React) | | apps/site | localdot.dev and the film | | packages/protocol | Types shared by the daemon and the app |

Run it from source: development guide.

Help

  • Open an issue with what you did and what localdot doctor says.
  • Other devices: Tailscale, an SSH tunnel, or the optional relay in Settings (remote access).
  • localdot uninstall removes localDot and keeps your data in ~/.localdot.

Licence

MIT. Third-party notices are in THIRD-PARTY.md.

Related Skills

View on GitHub
GitHub Stars10
CategoryAutomation
Updated23h ago
Forks1

Languages

TypeScript

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 low1 info