Skip to content

sequential-thinking.md

Latest commit

 

History

History
141 lines (116 loc) · 5.44 KB

sequential-thinking.md

File metadata and controls

141 lines (116 loc) · 5.44 KB
Raw

Sequential Thinking Tool

Multi-step reasoning tool for problems that need more than a quick thought. Each step should be a focused, concise observation or decision. Spread analysis across multiple steps rather than writing one long step.

Overview

Use sequential_thinking instead of think when:

  • The problem needs more than 1-2 sentences of reasoning
  • You need to revise or branch your analysis
  • The full scope is unclear and you may need course correction

The tool handles numbering, tracking, and branching automatically -- focus on your thinking content.

stdio transport only. This tool keeps thought history in the server process between calls. The HTTP transport is stateless, so the tool is not registered when running with --transport http.

Features

  • Auto-Managed State: Automatic thought numbering and progress tracking
  • Smart Revision Detection: Simply reference previous thoughts to revise
  • Simple Branching: Use explore labels for alternative approaches
  • Dynamic Adaptation: Tool adjusts to your thinking flow automatically
  • Context Preservation: Complete thought history maintained
  • Flexible Process: Add, revise, or explore thoughts naturally
  • Formatted Logging: Visual output shows thinking structure

Parameters

Required Parameters

  • action (string): Action to perform
    • Options: "think" (default) or "get_usage"
    • Description: Specifies whether to process a thought or get detailed usage instructions
  • thought (string): A single focused reasoning step
    • Description: 1-3 sentences covering one observation, decision, or question. Use multiple steps for longer analysis rather than one long step
    • Examples: "Let me break down this API design problem into authentication, rate limiting, and error handling"
  • continue (boolean): Whether more thinking is needed after this step
    • Description: true if you need more thoughts, false when done
    • Usage: Set to false only when satisfied with final answer

Optional Parameters

  • revise (string): Brief text snippet from previous thought to revise
    • Description: Reference text from a previous thought you want to reconsider
    • Examples: "API design problem", "authentication approach"
    • Effect: Automatically marks this as a revision and shows what you're revising
  • explore (string): Label for exploring alternative approach
    • Description: Identifier for branching into alternative reasoning path
    • Examples: "graphql-alternative", "security-first", "performance-focused"
    • Effect: Creates a named branch for this line of thinking

When to Use Sequential Thinking

Complex Problem Breakdown

When facing multi-faceted challenges requiring systematic analysis:

{
  "name": "sequential_thinking",
  "arguments": {
    "action": "think",
    "thought": "This API integration requires authentication, rate limiting, error handling, and data transformation. Let me work through each component systematically.",
    "continue": true
  }
}

Revising Previous Analysis

When new information contradicts earlier thinking:

{
  "name": "sequential_thinking",
  "arguments": {
    "action": "think",
    "thought": "Actually, my earlier assumption was wrong. The logs show connection pool exhaustion, not individual connection failures.",
    "continue": true,
    "revise": "database connections"
  }
}

Exploring Alternative Approaches

When considering different paths:

{
  "name": "sequential_thinking",
  "arguments": {
    "action": "think",
    "thought": "What if we used GraphQL instead of REST? This could simplify the client interface significantly.",
    "continue": true,
    "explore": "graphql-alternative"
  }
}

Final Decision

When reaching conclusion:

{
  "name": "sequential_thinking",
  "arguments": {
    "action": "think",
    "thought": "After comparing both approaches, REST with proper OpenAPI documentation offers better tooling support. Final design: REST API with OAuth2, rate limiting, and comprehensive validation.",
    "continue": false
  }
}

Best Practices

  1. Keep Steps Short: Aim for 1-3 sentences (~50-100 words) per step. Split longer reasoning into multiple steps
  2. Focus on Content: Concentrate on what you're thinking, not mechanics
  3. Natural Revision: Use revise when reconsidering previous thoughts
  4. Explore Alternatives: Use explore for different approaches
  5. Express Uncertainty: Natural uncertainty is valuable in thinking process
  6. Clear Conclusions: Set continue: false only with satisfactory answer

Advanced Usage

Getting Detailed Instructions

For comprehensive usage guide with examples:

{
  "name": "sequential_thinking",
  "arguments": {
    "action": "get_usage"
  }
}

Environment Variables

  • DISABLE_THOUGHT_LOGGING: Set to "true" to disable formatted console output
  • ENABLE_ADDITIONAL_TOOLS: Must include "sequential-thinking" to enable this tool

Key Advantages

  • Reduced Cognitive Load: No manual numbering or tracking
  • Natural Flow: Think naturally, tool handles structure
  • Smart Automation: Automatic state management and progress tracking
  • Flexible Process: Adapt thinking process dynamically
  • Clear Output: Formatted display shows thinking progression

The Sequential Thinking tool transforms complex problem-solving by removing mechanical overhead and letting you focus on the actual thinking process.