Deep Think

A Claude Code plugin for structured reasoning. Think through complex problems with strategies, branching, reflection, and persistent checkpoints.

Install

# Add the marketplace
/plugin marketplace add bis-code/mcp-deep-think

# Install the plugin
/plugin install deep-think@bis-code-mcp-deep-think

How It Works

Deep Think provides 5 MCP tools that work together:

strategize → think → reflect → branch → checkpoint
     |          |        |         |          |
  Pick a    Record    Check     Explore    Save/restore
  reasoning  steps    for       alternate  reasoning
  framework          issues    paths      state

Start a session with /start and describe what you're reasoning about. The plugin suggests a strategy and begins the thought chain.

Auto-checkpointing saves your reasoning before context compaction and every 10 thoughts (configurable). Resume anytime with /checkpoints.

Commands

| Command | What it does | |---------|-------------| | /start | Launch a reasoning session from a description | | /checkpoints | View, restore, or delete saved checkpoints | | /constraints | Manage reasoning rules and anti-patterns | | /practices | Manage review checklists and custom strategies |

Skills (auto-triggering)

| Skill | Triggers when... | |-------|-----------------| | restore-checkpoint | Starting a new session and continuing previous reasoning | | manage-constraints | User says "remember to always check X" or "add a rule about Y" | | manage-practices | User says "add a review step for..." or "create a strategy for..." |

Configuration

Create .deep-think.json in your project root to customize:

{
  "thinking": {
    "defaultStrategy": "first-principles",
    "maxThoughts": 50,
    "autoCheckpointEvery": 10
  },
  "practices": {
    "rules": ["Always consider security implications"],
    "antiPatterns": ["Premature optimization"],
    "reviewChecklist": ["Check for edge cases"]
  },
  "strategies": {
    "custom": [{
      "name": "incident-response",
      "description": "Structured incident investigation",
      "steps": ["Identify symptoms", "Gather evidence", "Form hypothesis", "Verify"],
      "checkpoints": ["Evidence collected?", "Root cause confirmed?"]
    }]
  }
}

Or use commands to manage configuration conversationally:

• /constraints — add rules like "always verify auth boundaries"
• /practices — add review steps or custom strategies

Built-in Strategies

| Strategy | Use when... | |----------|------------| | first-principles | Conventional wisdom may be misleading | | red-team | Validating important decisions | | convergent | Choosing between multiple options | | divergent | Brainstorming or stuck on a problem | | root-cause | Debugging or investigating failures | | decision-matrix | High-stakes decisions with multiple factors |

License

MIT