MCP servers / Sequential Thinking

Sequential Thinking MCP server: setup for Claude Code, Claude Desktop and Cursor

The Sequential Thinking server has a single tool, sequentialthinking, that gives the AI a structured way to work through a problem in numbered steps, revise earlier steps, and branch when it changes its mind.

It does not fetch anything or touch your files. It is a scratchpad with structure, and the server marks its one tool as read-only.

Runs with

npx (Node.js)

Package

@modelcontextprotocol/server-sequential-thinking on npm

Needs

Node.js (for npx).

Tools

1 (1 read-only)

Source

modelcontextprotocol/servers

Last checked

2026-09-20 (server reported version 2026.8.31)

Config for Sequential Thinking

Generated by the same code as the free builder, so it matches what you get there. Pick your client, then copy.

Claude Code

Save as .mcp.json in your project root. Claude Code asks you to approve it the first time.

{
  "mcpServers": {
    "thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ]
    }
  }
}

Or add it from a terminal instead (single quotes work in bash, zsh and PowerShell; more on this command):

claude mcp add thinking -- npx -y @modelcontextprotocol/server-sequential-thinking

Claude Desktop

Merge into claude_desktop_config.json (Settings → Developer → Edit Config), then fully quit and restart Claude Desktop.

{
  "mcpServers": {
    "thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ]
    }
  }
}

Cursor

Save as .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all projects.

{
  "mcpServers": {
    "thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ]
    }
  }
}

What it can do

These is the tool the server reported when we started it and asked on 2026-09-20. The labels come from the hints each tool declares about itself.

• sequentialthinking read-only
  A detailed tool for dynamic and reflective problem-solving through thoughts.

Watch out for

• It adds no new knowledge. It only shapes how the AI works through a problem, and models with built-in step-by-step reasoning may gain less from it. We have not measured the difference, so try it on a hard task before you keep it.
• The README says you can turn off logging of thought information by setting DISABLE_THOUGHT_LOGGING to true.
• It is safe to install, but every extra server adds tool descriptions the AI has to read, so remove it if you do not use it.

Variations

Turn off thought logging

From the README. The value is not a secret, so it is written directly in the config.

{
  "mcpServers": {
    "thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ],
      "env": {
        "DISABLE_THOUGHT_LOGGING": "true"
      }
    }
  }
}

The same in Claude Code, Claude Desktop and Cursor.

Try asking

• “Use sequential thinking to plan how to split this large function into smaller pieces.”
• “Think step by step about why this test only fails on CI, and revise if a step turns out to be wrong.”
• “Work through the trade-offs between these two database designs.”

Related servers

• Memory: A simple knowledge graph the AI can remember things in.
• Context7: Pulls up-to-date library docs into the conversation.