MCP 服务器

sequential-thinking-mcp

Remote MCP server for dynamic and reflective problem-solving through structured thinking. Enables step-by-step reasoning with branching, revision tracking, and session management.

README

🧠 Sequential Thinking MCP Server

A Remote MCP Server implementation for dynamic and reflective problem-solving through structured thinking. Built with the official @vercel/mcp-adapter for seamless Vercel deployment with Streamable HTTP transport.

✨ Features

🔄 Streamable HTTP Transport - Modern, efficient transport (recommended by MCP spec March 2025)
🌐 SSE Support - Backward compatibility with Server-Sent Events
📦 Vercel Native - One-click deployment with @vercel/mcp-adapter
🔀 Branching Logic - Explore alternative thinking paths
📝 Revision Tracking - Refine and improve previous thoughts
💾 Session Management - Persistent thinking across interactions
🏷️ Auto-Classification - Automatic thought type detection

🚀 Quick Deploy

Deploy to Vercel (Recommended)

After deployment, your MCP server will be available at:

Streamable HTTP: https://your-app.vercel.app/mcp
SSE: https://your-app.vercel.app/sse

Local Development

# Clone the repository
git clone https://github.com/sheikhcoders/sequential-thinking-mcp.git
cd sequential-thinking-mcp

# Install dependencies
npm install

# Build
npm run build

# Run stdio mode (for MCP clients)
npm start

# Run HTTP mode (for development)
npm run start:http

📡 Transport Modes

Streamable HTTP (Recommended)

The latest MCP transport specification. Eliminates persistent connections for better scalability.

POST https://your-app.vercel.app/mcp

Server-Sent Events (SSE)

Legacy transport for backward compatibility.

GET https://your-app.vercel.app/sse

Standard IO (stdio)

For local MCP clients like Claude Desktop.

node dist/index.js

🔧 Configuration

Claude Desktop

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "sequential-thinking": {
      "url": "https://your-app.vercel.app/mcp"
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "sequential-thinking": {
      "url": "https://your-app.vercel.app/mcp"
    }
  }
}

Cline (Streamable HTTP)

Add to cline_mcp_settings.json:

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": ["mcp-remote", "https://your-app.vercel.app/mcp"],
      "transportType": "Streamable HTTP"
    }
  }
}

For stdio (Local)

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": ["sequential-thinking-mcp"]
    }
  }
}

🛠️ Available Tools

`sequential_thinking`

The main tool for step-by-step problem solving with dynamic thought management.

Parameters:

Parameter	Type	Required	Description
`thought`	string	✅	Your current thinking step
`nextThoughtNeeded`	boolean	✅	Whether another thought is needed
`thoughtNumber`	number	✅	Current thought number (1-indexed)
`totalThoughts`	number	✅	Estimated total thoughts needed
`isRevision`	boolean	❌	Whether this revises previous thinking
`revisesThought`	number	❌	Which thought number is being revised
`branchFromThought`	number	❌	Create branch from this thought
`branchId`	string	❌	Branch identifier
`sessionId`	string	❌	Session ID for persistence

`get_thinking_summary`

Get a comprehensive summary of a thinking session.

`list_thinking_sessions`

List all available thinking sessions.

`switch_thinking_branch`

Switch between different thinking branches.

`complete_thinking_session`

Mark a session as completed with optional final conclusion.

📊 Thought Types

Thoughts are automatically classified:

Type	Detected When
`question`	Contains `?`, starts with what/how/why
`observation`	Contains "I notice", "I see", "observe"
`hypothesis`	Contains "perhaps", "maybe", "hypothesis"
`verification`	Contains "verify", "test", "check"
`insight`	Contains "insight", "realize", "aha"
`conclusion`	Contains "therefore", "in conclusion"
`refinement`	Contains "refine", "improve", "better"
`reflection`	Contains "reflect", "thinking about"
`analysis`	Default for analytical statements

🏗️ Project Structure

sequential-thinking-mcp/
├── api/
│   └── [transport]/
│       └── route.ts       # Vercel serverless handler
├── src/
│   ├── index.ts           # CLI entry point (stdio mode)
│   ├── server.ts          # MCP stdio server
│   ├── http-transport.ts  # Express HTTP server
│   ├── thinking-session.ts # Session management
│   └── types.ts           # TypeScript definitions
├── package.json
├── tsconfig.json
├── vercel.json            # Vercel configuration
└── README.md

🔒 Production Considerations

Session Persistence

For production with persistent sessions, integrate Vercel KV or Upstash Redis:

import { kv } from '@vercel/kv';

// Store session
await kv.set(`session:${sessionId}`, session);

// Retrieve session
const session = await kv.get(`session:${sessionId}`);

Authentication

Add OAuth or API key authentication for production:

// In your handler
const authHeader = request.headers.get('Authorization');
if (!validateToken(authHeader)) {
  return new Response('Unauthorized', { status: 401 });
}

📄 API Reference

Health Check

curl https://your-app.vercel.app/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "method": "initialize", "id": 1}'

Call Tool

curl https://your-app.vercel.app/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "sequential_thinking",
      "arguments": {
        "thought": "Let me analyze this problem step by step",
        "nextThoughtNeeded": true,
        "thoughtNumber": 1,
        "totalThoughts": 5
      }
    },
    "id": 1
  }'