🤖 AI-powered autonomous chat monitoring and management for Twitch streamers. Built on the Model Context Protocol (MCP) for seamless AI integration.
- AI-Powered Pattern Detection: Uses actual AI models (not keywords) to detect toxicity, spam, and engagement opportunities
- Intelligent Decision Making: Autonomously decides when to timeout, ban, engage with chat, or create polls
- Continuous Learning: Improves over time based on user feedback (1-5 star ratings)
- Comprehensive Logging: All actions and decisions logged in markdown for transparency
All the original Twitch MCP tools are included:
- Send messages to chat
- Create polls and predictions
- Generate clips
- Moderate chat (timeout/ban)
- Update stream title and category
- Analyze chat activity
- startAutonomousMonitoring - Start the AI agent
- stopAutonomousMonitoring - Stop and generate reports
- getAutonomousStatus - View current status and statistics
- orceAutonomousAnalysis - Force immediate analysis
- �ddUserFeedbackToAutonomous - Rate the AI's actions
- generateAutonomousReport - Generate performance reports
- Visit the Smithery server page (deployment pending)
- Click "Connect" to add to Cursor
- Configure with your Twitch credentials
- Start autonomous monitoring!
| Parameter | Description |
|---|---|
| witchClientId | Your Twitch application client ID |
| witchAuthToken | OAuth token (without 'oauth:' prefix) |
| witchBroadcasterId | Your Twitch user ID |
| witchChannel | Your Twitch channel name |
| Parameter | Default | Description |
|---|---|---|
| �utonomous.enabled | false | Enable autonomous monitoring |
| �utonomous.monitoringInterval | 30000 | Check interval in ms |
| �utonomous.confidenceThreshold | 0.7 | Min confidence for actions |
| �utonomous.spamDetection.enabled | true | Enable spam detection |
| �utonomous.toxicityDetection.enabled | true | Enable toxicity detection |
- Continuous Monitoring: AI monitors chat in real-time
- Pattern Analysis: Detects toxicity (with severity), spam, engagement opportunities
- Smart Decisions: AI decides which actions to take based on patterns
- Action Execution: Automatically executes timeouts, polls, engagement messages
- Feedback Loop: Streamers can rate actions to improve AI behavior
- Learning & Adaptation: System learns from feedback and adjusts over time
The AI agent creates detailed logs in markdown:
- eedback/actions-YYYY-MM-DD.md - Daily action log
- eedback/feedback-YYYY-MM-DD.md - Detailed feedback entries
- eedback/learning-insights.md - AI learning recommendations
- eedback/daily-report-YYYY-MM-DD.md - Daily performance reports
\\�ash
git clone https://github.com/YOUR_USERNAME/twitch-mcp-autonomous.git cd twitch-mcp-autonomous
npm install
npm run dev \\
The system is designed to work with any AI model. To connect your AI:
- Implement the AIAnalysisFunction interface
- Pass your AI function when initializing the autonomous monitor
- The system will use your AI for all pattern detection and decision making
Example AI prompts are provided for:
- Toxicity detection with severity scoring
- Spam identification
- Engagement opportunity detection
- Sentiment analysis
- Decision making
The autonomous system tracks:
- Total actions taken
- Success rate (based on feedback)
- Average user rating
- Most/least successful action types
- Pattern recognition accuracy
- Tool usage statistics
- Confidence Thresholds: Only acts on high-confidence decisions
- Cooldown Periods: Prevents spam and over-moderation
- Manual Override: Stop autonomous mode anytime
- Severity-Based Actions: Graduated responses based on severity
- Comprehensive Logging: Full transparency of all decisions
ISC
Built on the original Twitch MCP Server foundation, enhanced with autonomous AI capabilities.
Note: This is an enhanced version of the original twitch-mcp-smithery with autonomous AI features. For the basic version without AI monitoring, see the original repository.