Glen

Join the waitlist

We onboard teams in waves. Leave your work email and we'll send your invite when your spot opens.

Sequential Thinking MCP server

OfficialAnthropic (Model Context Protocol)86,565Config last verified Jun 1, 2026

A reference MCP server that gives agents a structured, revisable scratchpad for step-by-step reasoning.

Sequential Thinking is one of the reference servers maintained in the official Model Context Protocol repository. Rather than fetching external data, it gives the model a disciplined place to think: a single tool that records numbered thoughts, lets the agent revise earlier steps, branch into alternatives, and signal when more thinking is needed. The effect is a visible reasoning trace the model can build, audit, and correct mid-task instead of committing to a first guess.

It runs locally over stdio with no API key or external service, which makes it a safe, dependency-free addition to almost any agent. It pairs well with documentation and version-control servers: the model can pull facts from those tools, then use Sequential Thinking to plan how to apply them across a multi-step change before touching code.

Quick install

Copy-paste configs are provided for all 8 supported clients. Pick your client below.

Add to ~/.claude.json

~/.claude.json
json
{
  "mcpServers": {
    "sequential-thinking": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-sequential-thinking"
      ]
    }
  }
}
Or via CLI
bash
claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequential-thinking

Available tools

ToolDescription
sequentialthinkingRecords a single reasoning step in a flexible, reflective thinking process that can revise earlier thoughts, branch into alternatives, and adjust how many steps are needed.

What you can do with it

Plan a multi-file refactor

The agent decomposes a large change into ordered steps, revises the plan as it discovers constraints, and only then starts editing.

Work through ambiguous requirements

Branching thoughts let the model explore two interpretations of a task and pick the one that holds up before committing.

FAQ

Is it free?
Yes. It is an open-source reference server in the official MCP repo with no paid tier or external dependency.
Does it support remote/OAuth?
No. It runs locally over stdio and needs no authentication; there is no hosted remote endpoint.
Compare Sequential Thinking alternatives →← Browse all docs-and-context servers