Full-featured enough to ship real agents. Small enough to read over lunch.
🐍
Pure Python
Zero C-extensions, zero native binaries. If Python runs on your machine, MMClaw runs on your machine — laptop, server, or Raspberry Pi.
No build steps
⚡
1,700 Lines
The entire kernel fits in a single reading session. Step through the agent loop in your debugger. No magic, no black boxes.
Fully auditable
🔧
Kernel, Not App
A core engine built for extension. Swap providers, wire in custom tools, plug in your own state logic — in minutes, not hours.
Extensible
📡
Chat Connectors
Terminal, Telegram, Feishu, WhatsApp, QQ Bot — mix and match, same codebase. Long-polling throughout; no public IP, no webhooks needed.
Multi-connector
🌐
Browser Automation
Optional Playwright integration for real browser control — navigate, click, fill forms, scrape content, and maintain persistent login sessions across restarts. Enable via mmclaw config.
Playwright
🧠
Persistent Memory
Tell your agent to remember preferences, facts, or context — recalled automatically across all future sessions. Global memory persists beyond any single conversation.
Cross-session
🌐
Any LLM
OpenAI, Gemini, DeepSeek, Kimi, Claude, OpenRouter, and any OpenAI-compatible endpoint. Switch with a single config line.
Provider-agnostic
📖
Living Tutorial
The codebase is a teaching artifact. Learn the full agent loop and tool-calling protocol by reading real, production-grade Python.
Learn by reading