Skip to content
Conduit

mcp server

leuk

joyouslang/leukcommunitylistedunreviewed

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • uv (`curl -LsSf https://astral.sh/uv/install.sh | sh`)
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "leuk": {
      "command": "uvx",
      "args": [
        "leuk"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 26, 2026

leuk

A persistent AI agent that lives in your terminal.

Multi-provider LLMs · sub-agent orchestration · voice in/out · desktop & browser control · native multimodal (images, audio, video) · shell, files, web & MCP.

Getting Started · Documentation · Configuration · Architecture


Quickstart

uv sync          # install
uv run leuk      # start the REPL

The default provider is OpenCode Zen (big-pickle) — free, no API key, works out of the box. Want a different one? Run /auth inside the REPL to set up Anthropic, OpenAI, Google Gemini, OpenRouter, or a local Ollama/vLLM endpoint.

Everything else is configured from /settings (written to a single ~/.config/leuk/config.json) — no shell environment variables required.

❯ leuk
  leuk · zen/big-pickle                       ctx ~1.2k/200k 1%
❯ summarize today's git log and open a PR with the notes
  …

Why leuk

  • It remembers. Sessions are SQLite-backed, lazily created, auto-named, and resumable — pick up exactly where you left off.
  • It sees and hears. Screenshots, images, audio, and video are sent to the model natively (never base64-pasted as text); voice gives you hands-free speech in and out.
  • It does things. Run shell commands, read and patch files, fetch the web, drive a real browser, and control the desktop keyboard & mouse.
  • It stays safe. Rule-based review policies, per-tool permissions, persistent approvals, and channel-native approval prompts gate every risky action.
  • It scales out. Spawn sub-agents for parallel work, schedule background tasks, and reach it from Telegram, Slack, or Discord.

Highlights

ProvidersAnthropic (incl. OAuth), OpenAI, Google Gemini, OpenRouter, OpenCode Zen, local Ollama/vLLM — context window & capabilities are queried, never hardcoded
MultimodalImages, audio, and video sent through native provider blocks; non-vision models degrade gracefully
ToolsShell, file read/patch-edit, web fetch, sub-agents, memory, plus optional browser and desktop control
ExtensibleInstall SKILL.md agent skills (OpenClaw/ClawHub-compatible) and import MCP connectors from registries — manage with /skills and /mcp
VoiceLocal or remote STT & TTS with neural VAD; numbers and acronyms are spoken correctly per language
ContextTiered compaction (truncate → mask → summarize → drop) sized to each model's real context window
ReachMCP servers, a background scheduler, and channels (Telegram/Slack/Discord)

Documentation

The wiki is the authoritative, ground-truth documentation, kept in sync with the code:

Acknowledgments

leuk's offline voice stack builds on:

Citations

@misc{Silero Models,
  author = {Silero Team},
  title = {Silero Models: pre-trained text-to-speech models made embarrassingly simple},
  year = {2025},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/snakers4/silero-models}},
  email = {hello@silero.ai}
}

Silero model licenses & commercial use

leuk is AGPL-3.0. The Silero TTS models are downloaded at runtime (not bundled in this repo), so they do not alter leuk's license — but their own licenses constrain how you may use the synthesized output:

ComponentLicenseCommercial use
Silero VADMIT✅ yes
Silero TTS — CIS (v5_cis_base: ru, uk, kk, tt, uz, ky, ba, xal)MIT✅ yes
Silero TTS — other (v3_en, v3_de, v3_es, v3_fr, v4_indic)CC BY-NC❌ non-commercial only

So Russian and CIS-language TTS is free for commercial use; English, German, Spanish, French, and Indic TTS via Silero is non-commercial only. For commercial use of those languages, switch to the OpenAI TTS backend (or a permissively-licensed engine such as Piper). See Voice for details.

License

AGPL-3.0-or-later

↑↓ navigate openesc close