Think MCP is a structured reasoning server that enables LLMs to break down complex problems through sequential thinking, branching, and cross-session memory.
Core Capabilities:
Sequential Reasoning - Use think to add step-by-step thoughts with confidence scoring (1-10), sub-steps breakdown, and alternative approaches. Use think_batch to submit up to 30 thoughts in a single call for high-velocity reasoning with atomic validation.
Branching & Revision - Explore alternative solution paths from any thought using branchId or revise previous thoughts with isRevision: true to correct mistakes without losing context.
Deep Analysis Framework - Generate structured methodologies with think_logic through 4 phases: Chain Mapping → Crack Hunting → Standard Benchmark → Action Planning. Customize by depth (quick/standard/deep), focus (security, performance, reliability, UX, data-flow), and technology stack (NestJS, React, Redis, Next.js, etc.).
Cross-Session Memory - Search current session thoughts or retrieve insights from previous sessions (24-hour retention) using think_recall with fuzzy matching across thoughts, extensions, and alternatives.
Session Management - Finalize with think_done to validate logic gaps, unresolved blockers, and low-confidence thoughts, with optional export in Markdown or JSON format (including Mermaid diagrams). Use think_reset to clear context and start fresh.
Intelligent Assistance - Built-in Nudge System provides proactive warnings for low confidence, tunnel vision, missing breakdowns, and unresolved blockers, plus complexity-based tool recommendations and stagnation detection.
Performance Features - ~55% token reduction through imperative prompts, ASCII tree visualization, quick inline extensions (critique, elaborate, correct, innovate, polish), and progress tracking with next-action hints.
Supports exporting thought sessions in Markdown format for human-readable documentation of reasoning processes.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Think MCPhelp me debug this authentication error in our NestJS API"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
🧠 Think MCP
Structured Sequential Thinking Server for LLMs
Quick Start • Tools • Features • Changelog
📖 Overview
Think MCP transforms how LLMs approach problem-solving. It's not just a tool; it's a cognitive framework that enables:
Sequential Reasoning: Step-by-step problem decomposition.
Branching & Revision: Ability to backtrack, fork thoughts, and correct mistakes.
Deep Analysis: Built-in methodology generator for rigorous code and logic auditing.
Long-term Memory: Cross-session recall of insights and dead ends.
✨ Features
Feature | Description |
🚀 Efficient Thinking | Imperative Prompts (v5.1) reduce token usage by ~55% using IF/THEN logic. |
⚡️ Burst Mode | Submit up to 30 thoughts in a single API call with |
🧠 Methodology Generator | On-demand deep analysis frameworks (Chain Mapping, Crack Hunting, etc.) via |
💾 Smart Memory | Cross-session learning via |
🔔 Nudge System | Proactive micro-prompts to detect low confidence, tunnel vision, or missed steps. |
🌳 Branching | Explore alternative paths without losing context. |
🚀 Quick Start
Installation
MCP Configuration
Add this to your MCP settings file:
🛠️ Tools Reference
1. think
The core unit of reasoning. Adds a single thought to the chain.
2. think_batch
High-velocity reasoning. Submit a complete chain (1-30 thoughts) at once.
⚠️ Constraints
IF similarity > 60%→ Reject ("Stagnation")
IF thought < 50 chars→ Reject ("Too short")
IF avg_confidence < 4→ Warning issued
3. think_logic
The Architect. Generates a strict thinking methodology for complex analysis.
Output Phases: CHAIN MAPPING → CRACK HUNTING → STANDARD BENCHMARK → ACTION PLANNING
4. think_recall
The Memory Bank. Search current session or past insights.
Best Practices:
BEFORE complex_task→ Checkscope: 'insights'IF repeating_logic→ Check for dead endsIF unsure→ Verify context
5. think_done & think_reset
think_done: Finalize session. Validates gaps, blockers, and confidence levels.think_reset: Wipe the slate clean. (Use only if task context changes completely).
💡 Intelligent Systems
The Nudge System
The server watches your back.
Trigger Pattern | System Nudge |
| "Low confidence. Validate?" |
| "No alternatives. Tunnel vision?" |
Complex goal w/o subSteps | "Complex goal, no breakdown. Decompose?" |
Unresolved blocker | "Blocker unresolved. Fix first." |
Complexity Budget
Recommended tool usage based on task size.
Task Difficulty | Thoughts | Recommended Tool |
Simple | 0-2 | Skip (Direct Answer) |
Medium | 3-7 |
|
Complex | 8-30 |
|
🔄 Changelog
Imperative Prompts: Switched to IF/THEN style instructions.
Performance: ~55% Token Reduction per request.
Optimization: Faster parsing, less LLM overhead.
New Tool:
think_logicfor generating methodologies.Framework: Added 4-phase analysis (Mapping, Cracking, Benchmarking, Planning).
v4.6.0: Added NudgeService for proactive prompts.
v4.5.0: Renamed to
think, added Cross-session insights.v4.1.0: Introduced Burst Thinking (
think_batch).