S
Official Memory & Knowledge

Sequential Thinking

Anthropic's official structured thinking and reasoning helper.

Works with: Claude DesktopClaude CodeCursorWindsurfClineVS Code (Continue)Zed
Quick install
npx -y @modelcontextprotocol/server-sequential-thinking

How to install the Sequential Thinking MCP server

Add this to your Claude Desktop MCP configuration:

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

Add this to your Claude Code MCP configuration:

npx -y @modelcontextprotocol/server-sequential-thinking

Add this to your Cursor MCP configuration:

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

Add this to your Windsurf MCP configuration:

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

Add this to your Cline MCP configuration:

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

Add this to your VS Code (Continue) MCP configuration:

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

Add this to your Zed MCP configuration:

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

The Sequential Thinking MCP server is Anthropic’s official structured-reasoning helper. It gives Claude a dedicated scratchpad with revise-and-branch primitives, turning free-form chain-of-thought into something explicitly inspectable.

This is one of the more conceptual MCP servers. It doesn’t connect to an external system. It just gives Claude a better mental model for solving multi-step problems. The result is noticeably better reasoning on complex tasks.

Why use it

Default Claude reasoning is good but unstructured. Long chains of thought get lost; revisions overwrite earlier ideas; branching becomes implicit. The Sequential Thinking server makes all three explicit.

For Claude Code workflows where you’re debugging something tricky, this server pays for itself. Claude can think “let me check assumption X” → check it → “that was wrong, let me revise” → branch into a new direction. The conversation transcript becomes a debuggable artefact.

What it actually does

One tool: think. Each call records a thought, an optional revision-of, an optional branch-from, and a flag for whether more thinking is needed. Claude calls this tool repeatedly across a hard problem.

Practical patterns:

  • Complex debugging: Claude breaks the problem into hypotheses, tests each, revises when wrong.
  • Architecture decisions: Claude lays out options, weighs trade-offs explicitly, branches into details.
  • Long research: Claude tracks what it has and hasn’t checked, marking each thread.

Gotchas

The server doesn’t enforce good thinking. If you don’t prompt for it, Claude won’t always reach for the tool. Adding “use sequential thinking to break this down” to your prompt nudges it.

The thought log lives in the conversation, not on disk. If you close the conversation, the reasoning is gone. For persistent reasoning artefacts, copy the relevant thoughts into Memory or Notion before closing.

Pair with the Time server for any workflow that involves dates or scheduling. Sequential Thinking gives Claude the scratchpad; Time gives it accurate temporal grounding.

Also in Memory & Knowledge

Combine Sequential Thinking with ContextBolt

Sequential Thinking gives Claude one kind of memory. ContextBolt adds another: every tweet, post, and article you save across X, Reddit, and LinkedIn becomes searchable by meaning. Run both as MCP servers and Claude can pull from both layers in one prompt.

See ContextBolt →

MCP servers related to Sequential Thinking

All in Memory & Knowledge →

ContextBolt

Search your X, Reddit, and LinkedIn bookmarks from inside any MCP client.
M

Memory

Anthropic's official knowledge-graph memory server with persistent entity storage.
m

mem0

Open-source memory layer with vector search across past interactions.
P

Pieces

Long-term memory for developers, with snippet capture and IDE integration.

Sequential Thinking MCP server: FAQs

What does sequential thinking actually do?

It exposes a single tool that lets Claude record a thought, mark whether more thinking is needed, optionally revise an earlier thought, or branch into a new line of reasoning. The server stores the thought sequence and surfaces it back to Claude on demand.

Does it use a separate model?

No. The server doesn't call any model. It's a structured scratchpad that Claude itself fills in. The benefit is that Claude treats the thoughts as first-class data rather than free-form prose.

When should I install it?

For any workflow involving complex multi-step planning, debugging, or research. It pairs well with the GitHub and Filesystem servers for engineering work, where Claude can think through a fix step by step before writing code.

Does it persist across conversations?

No. Each conversation starts fresh. For persistent memory, combine it with the Memory MCP server.

How is it different from Claude's normal reasoning?

Without the server, Claude reasons in plain text. With it, reasoning becomes a structured tool call: Claude can explicitly mark a thought as 'revising thought 3' or 'branching from thought 7'. This produces cleaner, more auditable thinking.

Back to all MCP servers