MCP Directory

Sequential Thinking

Official

Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.

Verified
stdio (local)
No auth
TypeScript

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run
npx -y @modelcontextprotocol/server-sequential-thinking

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

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

Before you start

  • Node.js with npx (for the npm package), OR
  • Docker installed and running (to use the mcp/sequentialthinking image)
  • No API key or network access required

About Sequential Thinking

Sequential Thinking is a reference MCP server from the official modelcontextprotocol/servers repo. It exposes a single tool, sequentialthinking, that gives a model a structured scratchpad for breaking a hard problem into a numbered sequence of thoughts — with the ability to revise earlier steps, branch into alternatives, and dynamically grow or shrink the total number of steps as understanding improves.

It holds no external state and needs no credentials or network access. Each call records one thought (its number, the running total estimate, and flags like isRevision, revisesThought, branchFromThought, and nextThoughtNeeded) and the model keeps calling until nextThoughtNeeded is false. This nudges the model toward explicit, auditable reasoning rather than one-shot answers.

It's purely a reasoning aid — it does not call other tools or fetch data. Use it for planning, analysis, and multi-step decomposition where you want the chain of thought to be visible and revisable.

Tools & capabilities (1)

sequentialthinking

Record one reasoning step with thought, thoughtNumber, totalThoughts, and nextThoughtNeeded; supports optional isRevision/revisesThought, branchFromThought/branchId, and needsMoreThoughts for revision and branching

When to use it

  • Use it when a task needs multi-step planning and you want the model to lay out steps explicitly
  • Use it when the model may need to revise an earlier assumption partway through a problem
  • Use it when exploring alternative approaches via branching before committing to one
  • Use it when the scope is unclear up front and the number of reasoning steps may change
  • Use it when you want a visible, auditable chain of reasoning rather than a single opaque answer

Quick setup

  1. 1Add the server to your MCP client config using command `npx` with args `["-y", "@modelcontextprotocol/server-sequential-thinking"]` (or use the Docker image `mcp/sequentialthinking`)
  2. 2On Windows, wrap the npx call with `cmd /c` if needed
  3. 3Optionally set DISABLE_THOUGHT_LOGGING=true to suppress thought logging to stderr
  4. 4Restart your MCP client so it launches the server
  5. 5Verify by asking the model to reason through a multi-step problem and confirm it calls the sequentialthinking tool

Security notes

Runs entirely locally with no network access or stored data, so the attack surface is minimal. The only risk is the usual one of executing an npm package, so pin to the official scope.

Sequential Thinking FAQ

Does it need an API key or internet access?

No. It is a self-contained local reasoning tool — no credentials, no network calls, no external services.

What tool does it actually expose?

A single tool, `sequentialthinking`, which records one reasoning step per call with parameters like thoughtNumber, totalThoughts, and nextThoughtNeeded, plus optional revision and branching fields.

How do I stop it from printing thoughts to the console?

Set the environment variable DISABLE_THOUGHT_LOGGING=true in the server's env block to disable the formatted thought logging.

npx vs Docker — which should I use?

Either works. Use npx for the quickest local setup if you have Node; use the `mcp/sequentialthinking` Docker image for an isolated, dependency-free run.

Does it run other tools or fetch data?

No. It only structures the model's own reasoning. It does not browse, call APIs, or invoke other MCP servers.

Alternatives to Sequential Thinking

Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.

Verified
stdio (local)
No auth
TypeScript
9 tools
Updated 5 months agoRepo

Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.

Verified
HTTP (remote)
No auth
Hosted
6 tools
Updated 5 months agoRepo

Official Qdrant server using a vector collection as semantic memory: store and find embeddings.

Verified
stdio (local)
API key
Python
2 tools
Updated 1 month agoRepo