CoT Reasoning Map

Learn from Claude's chain of thought. An MCP App (SEP-1865) for Claude Code that turns extended-thinking traces into interactive reasoning maps — typed moves (framing, hypothesis, verification, backtrack, self-correction, insight, action, conclusion) laid out as a graph with dependency and discard edges, an effort timeline, replay mode, and click-through to the verbatim CoT excerpts.

No API key needed: the Claude session you're already running does the annotation.

How it works

1. list_sessions — finds recent Claude Code transcripts (~/.claude/projects/**/*.jsonl) containing thinking blocks
2. get_thinking — extracts the numbered thinking blocks + tool-call skeleton from a session
3. The calling Claude annotates the thinking into typed moves with verbatim excerpts
4. show_reasoning_map — renders the moves as an interactive widget inline in the chat (Claude Code desktop app / claude.ai; text fallback in the terminal CLI)

Say things like: "map the reasoning of my last session" or "show me a demo reasoning map" (no-args call renders sample data).

Install

Requires Node 18+.

git clone https://github.com/matanma12/cot-widget-spike.git
cd cot-widget-spike
npm install && npm run build
claude mcp add cot-spike --scope user -- node "$(pwd)/server.js"

Restart Claude Code and ask for a reasoning map.

Development

• server.js — MCP server (stdio; npm run serve:http for Streamable HTTP on :3001)
• mcp-app.html + src/mcp-app.js — the widget (App from @modelcontextprotocol/ext-apps)
• npm run build — bundles to dist/mcp-app.html (vite-plugin-singlefile)
• node test-client.js — protocol smoke test over stdio

Status / roadmap

Working: ingestion, host-model annotation loop, graph + timeline + replay + excerpts, verified inline rendering in Claude Code desktop. Next: live session following via hooks, pattern library across traces, predict-the-next-move replay, claude.ai deployment.

MIT licensed.