Why Thunder Thinking?

The official sequential-thinking MCP server gives you a basic loop: thought in, thought out, next. That's it. No memory of what you thought, no confidence in your conclusions, no ability to branch or revise, no persistence across sessions.

Thunder Thinking fixes all of that:

| Feature | sequential-thinking | thunder-thinking | |---------|:--------------------:|:-----------------:| | Basic sequential thoughts | Yes | Yes | | Typed thought categories | No | 9 types (analysis, hypothesis, verification, revision, decision, decomposition, synthesis, critique, observation) | | Confidence scoring | No | 0.0 - 1.0 per thought | | Branching | No | Branch from any thought | | Revision tracking | No | Mark which thought you're revising | | Chain persistence | No | Saved to disk as JSON | | Chain summaries | No | Type distribution, avg confidence, duration | | Clear and restart | No | Old chains preserved on clear | | Session isolation | No | Per-session chain IDs |

Install

npm install -g thunder-thinking

Or clone and build locally:

git clone https://github.com/Orellius/thunder-thinking.git
cd thunder-thinking
npm install
npm run build

Quick Start

With Claude Code

claude mcp add thunder-thinking node ~/path/to/thunder-thinking/dist/index.js

With Cursor / VS Code

{
  "mcpServers": {
    "thunder-thinking": {
      "command": "node",
      "args": ["/path/to/thunder-thinking/dist/index.js"]
    }
  }
}

Tools

sequentialthinking

The core reasoning tool. Each thought includes:

• thought — your current thinking step
• thoughtNumber / totalThoughts — position and estimated total (adjustable)
• thoughtType — one of: analysis, hypothesis, verification, revision, decision, decomposition, synthesis, critique, observation
• confidence — 0.0 to 1.0 confidence in this thought's conclusion
• isRevision / revisesThought — mark revisions explicitly
• branchFromThought / branchId — create alternative reasoning paths
• needsMoreThoughts — signal that the chain should continue
• tags — free-form labels for categorization

Returns: chain ID, session ID, average confidence, type distribution, thought history length.

thinking_clear

Clear the current chain and start fresh. The old chain is preserved on disk for later review.

thinking_summary

Get a summary of the current chain: total thoughts, branches, revisions, average confidence, type distribution, duration, and all thoughts with their metadata.

How It's Different

Before (sequential-thinking):

Thought 1: "Let me think about this..."
Thought 2: "Actually, maybe..."
Thought 3: "I think the answer is X"
// Gone. No record. No confidence. No structure.

After (thunder-thinking):

Thought 1 [analysis, confidence: 0.7]: "Let me break this into components..."
Thought 2 [hypothesis, confidence: 0.6]: "If we approach it this way..."
Thought 3 [verification, confidence: 0.4]: "Wait, that assumption is wrong"
Thought 4 [revision of #2, confidence: 0.8]: "Better approach: ..."
Thought 5 [decision, confidence: 0.92]: "Final answer with high confidence"
// Persisted. Typed. Scored. Reviewable.

Chain Persistence

Every chain is saved to ~/.thunder-thinking/chains/ as JSON. You can:

• Review past reasoning after the session ends
• Compare chain quality across different approaches
• Audit decision-making in production AI agents

Requirements

• Node.js: 18+
• MCP client: Claude Code, Cursor, Windsurf, or any MCP-compatible tool

License

MIT — use it however you want.