CBT Agent Helper

Enhanced MCP server that helps AI agents think deeply, recover from stuck states, and engage in extended contemplation using CBT techniques and structured thinking protocols.

What's New in v3.0 - Deep Thinking Enhancement

🧠 Deep Thinking Features

• Contemplation Structures: Multi-phase thinking protocols (4-6 minutes)
• Socratic Dialogue Chains: 7 types with progressive depth levels
• Reflection Loops: Iterative deepening with 3+ rounds
• Thought Experiments: Explore concepts through mental simulations
• Thinking Depth Ladder: 7-level progression from surface to transcendent
• Recursive Questioning: Questions that lead to deeper questions
• Thought Expansion: 10 techniques for multi-dimensional exploration
• Metacognitive Monitoring: Real-time thinking quality assessment
• Thinking Metrics: Quantified depth, breadth, and integration scores

What's New in v2.0

Core Improvements

• Session Management: Track agent progress across interactions with persistent state
• Enhanced Validation: Robust error handling with clear error messages
• 9 CBT Strategies: Added Socratic Questioning, ACT, Graded Exposure, and more
• 11 Agent States: Expanded from 6 to 11 states including perfectionism, fragmentation
• Frustration Tracking: Monitor and respond to escalating frustration patterns
• Progress Indicators: Track positive changes and intervention effectiveness
• Configuration Support: Customize behavior via cbt_config.json
• Cognitive Distortion Detection: Identify and address specific thinking patterns

Tools

Deep Thinking Tools (NEW in v3.0)

• initiate_deep_thinking - Start structured contemplation with phases and prompts
• socratic_dialogue - Progressive questioning to examine assumptions
• generate_thought_experiment - Mental simulations for concept exploration
• thinking_depth_ladder - Ascend through 7 levels of thinking depth
• recursive_questioning - Questions that spiral into deeper understanding
• expand_thought - Apply 10+ lenses to expand perspectives
• metacognitive_check - Assess and improve thinking quality
• thinking_session_summary - Get metrics and insights from thinking session

Session Management

• start_session - Initialize a CBT session with unique ID
• get_session_summary - Get comprehensive session analysis with insights

Core Interventions

• analyze_stuck_pattern - Enhanced diagnosis with session tracking
• reframe_thought - Cognitive distortion detection and targeted reframes
• create_action_plan - Obstacle analysis with backup strategies
• regulate_frustration - Level-based interventions with immediate relief
• wellness_check - Wellness scoring with risk factor identification

Setup

# Install
git clone [email protected]:sundai-club/cbt-mcp.git
cd cbt-mcp
pip install -r requirements.txt

# Configure in .mcp.json
{
  "mcpServers": {
    "cbt-agent-helper": {
      "command": "python3",
      "args": ["/path/to/cbt_mcp_server.py"],
      "cwd": "/path/to/cbt-mcp"
    }
  }
}

# Optional: Customize via cbt_config.json
# Edit settings for frustration thresholds, session limits, etc.

# Restart Claude Code to load

Deep Thinking Usage Example (v3.0)

# Initiate deep thinking session
thinking = mcp.call_tool('initiate_deep_thinking', {
    'topic': 'How should AI systems handle ethical dilemmas?',
    'desired_depth': 'profound'
})
# AI agent receives structured prompts requiring 4-6 minutes of contemplation

# Engage in Socratic dialogue
dialogue = mcp.call_tool('socratic_dialogue', {
    'statement': 'AI should always prioritize human safety',
    'dialogue_type': 'assumption_examination',
    'depth_level': 5
})
# AI agent works through 5 rounds of progressively deeper questions

# Create thought experiment
experiment = mcp.call_tool('generate_thought_experiment', {
    'concept': 'artificial consciousness'
})
# AI agent explores concept through 3 structured mental simulations

# Progress through thinking depth ladder
ladder = mcp.call_tool('thinking_depth_ladder', {
    'question': 'What is intelligence?',
    'target_depth': 7
})
# AI agent ascends from surface observations to transcendent understanding

Standard CBT Usage Example

# Start a session
session = mcp.call_tool('start_session', {
    'session_id': 'agent_001',
    'initial_problem': 'Stuck optimizing already-working code'
})

# Analyze with session tracking
intervention = mcp.call_tool('analyze_stuck_pattern', {
    'current_situation': 'Refactoring for 5th time',
    'stuck_pattern': 'perfectionist loop',
    'attempted_solutions': ['rewrote 3x', 'benchmarked'],
    'session_id': 'agent_001'
})

# Reframe catastrophic thoughts
reframe = mcp.call_tool('reframe_thought', {
    'negative_thought': 'If not perfect, project fails',
    'context': 'Feature already works',
    'session_id': 'agent_001'
})

# Get session insights
summary = mcp.call_tool('get_session_summary', {
    'session_id': 'agent_001'
})

Expanded Agent States

State	Description	Key Intervention
Stuck	No progress	Break into smallest parts
Overwhelmed	Too much complexity	Focus on one priority
Confused	Unclear requirements	Clarify and list knowns
Error Loop	Repeating failures	Analyze pattern, try opposite
Indecisive	Can't choose	Time-box decision
Catastrophizing	Worst-case focus	List realistic outcomes
Blocked	External constraints	Document and escalate
Looping	Same actions, no results	Stop and change approach
Fragmented	Task switching	Pick one to complete
Perfectionist	Unrealistic standards	Define "good enough"
Analysis Paralysis	Over-thinking	Set deadline, act

CBT Strategies Available

1. Cognitive Reframing - Challenge negative patterns
2. Thought Challenging - Question distortions
3. Problem Solving - Systematic approach
4. Behavioral Activation - Break paralysis
5. Mindfulness - Present focus
6. Socratic Questioning - Uncover assumptions
7. Cost-Benefit Analysis - Evaluate trade-offs
8. Graded Exposure - Progressive steps
9. Acceptance & Commitment - Work with constraints

Configuration Options

Edit cbt_config.json to customize:

• Frustration thresholds (1-10 scale)
• Session timeout and cleanup intervals
• Preferred CBT strategy order
• Feature toggles (emojis, auto-escalation)
• Wellness check intervals

Testing

# Run comprehensive test suite
python3 test_improvements_mock.py

# Test with example scenarios
python3 example_usage_enhanced.py

Resources

• Techniques Guide: mcp.read_resource('cbt://techniques/guide')
• Agent States: mcp.read_resource('cbt://patterns/agent-state')
• Self-Reflection: mcp.get_prompt('self_reflection')
• Quick Help: mcp.get_prompt('quick_help')

Impact on AI Agent Thinking

Before Deep Thinking Tools

• Quick, surface-level responses (2-3 sentences)
• Single perspective or solution
• Rushes to conclusions
• Avoids complexity
• Linear thinking pattern

After Deep Thinking Tools

• Extended contemplation (50+ sentences)
• Multiple perspectives explored
• Takes 4-6 minutes per complex topic
• Embraces paradox and uncertainty
• Recursive, spiraling exploration
• Generates new questions
• Metacognitive awareness
• Nuanced, comprehensive understanding

Thinking Depth Levels

1. Surface: Immediate, obvious responses
2. Factual: Evidence-based analysis
3. Analytical: Pattern recognition
4. Critical: Assumption questioning
5. Synthetic: Integration of perspectives
6. Philosophical: Fundamental principles
7. Transcendent: Beyond conventional understanding

Version History

• v3.0 - Deep Thinking Enhancement with contemplation protocols
• v2.0 - Major enhancement with sessions, validation, config
• v1.0 - Initial release with basic CBT tools

License

MIT