mcp server

qanot

The AI agent that flies on its own. Two commands to fly. 🪶

QANOT/qanotcommunitylistedunreviewed

Use tool

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": {
    "qanot": {
      "command": "uvx",
      "args": [
        "qanot"
      ]
    }
  }
}

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 22, 2026

Qanot (Uzbek for "wing") — lightweight Python framework for AI-powered Telegram agents with memory, tools, RAG, multi-agent delegation, and multi-provider failover. All out of the box.

Quick Start

pip install qanot
qanot init        # Interactive setup — picks provider, model, starts bot

That's it. The wizard configures everything and auto-starts your bot.

What Makes Qanot Different

Agent Loop — Up to 25 tool-use iterations per turn with circuit breaker, result-aware loop detection, and smart error recovery.

3-Tier Model Routing — Routes messages by complexity. "salom" -> Haiku ($0.003). "REST API yozib ber" -> Opus ($0.029). Saves 50-60% on costs.

1M Context Window — Auto-detects 1,000,000 tokens for Claude Opus/Sonnet 4.6. Server-side compaction for infinite conversations.

22 Telegram Commands — Full settings management from chat: /model, /think, /voice, /lang, /mode, /routing, /mcp, /plugins, /usage, /context, /export, and more. All with inline keyboard buttons.

Multi-Agent — Agents that talk to each other. Delegate tasks, hold conversations between agents, spawn background workers.

5 Providers — Claude, GPT, Gemini, Groq, Ollama. Automatic failover with smart cooldowns.

Anthropic Enhancements — Server-side code execution (free sandbox), trained memory tool, thinking.display: "omitted" for faster streaming. All auto-injected for Claude.

3-Tier Memory — WAL captures real-time corrections. Daily notes log conversations. Long-term memory in MEMORY.md. Anthropic memory tool (/memories). Agent evolves its own SOUL.md over time. RAG indexes everything.

RAG — Hybrid search (vector + FTS5) with FastEmbed CPU embedder. No GPU needed.

Voice — 4 providers (Muxlisa, KotibAI, Aisha, Whisper). STT + TTS. Uzbek native.

Streaming — Native Telegram sendMessageDraft (Bot API 9.5). Real-time, not edit-message hack.

MCP Client — Connect to 1000+ community MCP servers (filesystem, Postgres, GitHub, etc.). pip install qanot[mcp]

Browser Control — Browse, click, fill forms, screenshot via Playwright. pip install qanot[browser]

Skills System — Agent creates reusable skills (SKILL.md + scripts). Hot-reload without restart.

Plugins — amoCRM, Bitrix24, 1C Enterprise, AbsMarket, iBox POS, and more. Build your own with @tool decorator.

Security — 3-tier exec security (cautious default), per-user rate limiting, SSRF protection, safe file writes, SecretRef.

Architecture

User -> Telegram -> Agent Loop (25 iterations max)
                      |-- Model Router (Haiku / Sonnet / Opus)
                      |-- LLM Provider (Claude / GPT / Gemini / Groq / Ollama)
                      |-- Tool Registry (40+ built-in + plugin tools)
                      |-- Memory (WAL -> daily notes -> MEMORY.md -> /memories)
                      |-- RAG Engine (FastEmbed + FTS5 hybrid)
                      |-- Voice Pipeline (4 providers)
                      |-- Multi-Agent (delegate / converse / spawn)
                      |-- MCP Client (external tool servers)
                      |-- Browser (Playwright headless)
                      |-- Skills (hot-reloadable scripts)
                      |-- WebChat (WebSocket adapter)
                      |-- Webhook (external event handler)
                      +-- Security (rate limit + exec approval + file jail)

Telegram Commands

All commands have inline keyboard buttons for easy settings management:

Category	Commands
Settings	/model, /think, /voice, /voiceprovider, /lang, /mode, /routing, /group, /exec, /code
Info	/status, /usage, /context, /config, /mcp, /plugins, /id
Actions	/reset, /compact, /export, /stop
Help	/help

CLI

qanot init                  # Setup wizard
qanot start / stop / restart
qanot status / logs
qanot update                # Self-update from PyPI
qanot doctor --fix          # Health check + auto-repair
qanot config show / set
qanot plugin install <name> # Install from registry
qanot plugin new <name>     # Scaffold a plugin

Plugins

Ready-made integrations at QanotHub:

Plugin	What it does
amoCRM	Leads, contacts, companies, tasks, chats, tags, pipelines
Bitrix24	Deals, leads, contacts, invoices, quotes, products, tasks
1C Enterprise	Contractors, products, sales, purchases, cash, balances
AbsMarket	POS sales, purchases, inventory, customers, suppliers
iBox POS	Products, stock, orders, payments, analytics, ABC analysis

Build your own:

from qanot.plugins.base import Plugin, tool

class QanotPlugin(Plugin):
    name = "my_plugin"

    @tool("my_tool", "What this tool does")
    async def my_tool(self, params: dict) -> str:
        return '{"result": "done"}'

Docs

	English	O'zbekcha
Getting Started	docs	docs
Full Guide	docs	docs
Configuration	docs	docs
Tools	docs	docs
Plugins	docs	docs
Architecture	docs	docs
API Reference	docs	docs

Contributing

git clone https://github.com/QANOT/qanot.git
cd qanot
pip install -e .
python -m pytest tests/ -v   # 1007 tests

License

MIT — use it, fork it, build with it.