# Sequential Thinking

> Structured step-by-step reasoning for complex problem solving.

[Canonical HTML page](https://top-mcps.com/mcp/sequential-thinking) · [server.json](https://top-mcps.com/mcp/sequential-thinking.json) · [methodology](https://top-mcps.com/about/methodology)

## Install

### Claude Desktop — `claude_desktop_config.json`

Paste under mcpServers. Fully quit and reopen Claude after editing.

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

### Claude Code — `CLI or .mcp.json`

Run from your repo. Commit .mcp.json to share with your team.

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

### Cursor — `.cursor/mcp.json`

Global path: ~/.cursor/mcp.json. Reload window after editing.

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

### VS Code — `.vscode/mcp.json`

VS Code uses the "servers" key (not "mcpServers").

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

### Windsurf — `~/.codeium/windsurf/mcp_config.json`

Open via Cascade → hammer icon → Configure.

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

### Cline — `cline_mcp_settings.json`

Open via the Cline sidebar → MCP Servers → Edit.

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

### Continue — `~/.continue/config.json`

Continue uses modelContextProtocolServers with a transport block.

```json
{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "npx",
          "args": [
            "-y",
            "@modelcontextprotocol/server-sequential-thinking"
          ]
        }
      }
    ]
  }
}
```

### Codex CLI — `~/.codex/config.toml`

Codex uses TOML. Each server is a [mcp_servers.<name>] subtable.

```shell
# ~/.codex/config.toml
[mcp_servers.sequential-thinking]
command = "npx"
args = [
  "-y",
  "@modelcontextprotocol/server-sequential-thinking",
]
```

### Zed — `~/.config/zed/settings.json`

Zed calls them "context_servers". Settings live-reload on save.

```jsonc
{
  "context_servers": {
    "sequential-thinking": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "@modelcontextprotocol/server-sequential-thinking"
        ]
      }
    }
  }
}
```

### ChatGPT — `ChatGPT → Apps directory`

Sequential Thinking doesn't ship a hosted HTTPS endpoint today. ChatGPT supports remote MCP servers only — to use this server in ChatGPT you'll need to deploy it to a public HTTPS URL first (e.g. via Cloudflare Workers or Vercel) or wait for an official remote build.

```none

```

## At a glance

- **Maintainer:** Anthropic (modelcontextprotocol)
- **Transport:** stdio
- **Auth model:** None
- **Required secrets:** None
- **Supported clients:** Claude, Any MCP-compatible client
- **License:** MIT
- **Language:** TypeScript
- **Latest version:** latest
- **Last verified:** 2026-05-31
- **GitHub stars:** 84,102 (fetched 2026-06-02T11:55:51.496Z)
- **Score:** 91/100 (rubric 2026-04 — see https://top-mcps.com/about/methodology)
- **Source:** https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking

## Tools & permissions

| Tool | Description | Args | Side effects |
|------|-------------|------|--------------|
| `sequentialthinking` | Record one step of a structured reasoning chain and optionally revise previous steps. | `thought: string, thoughtNumber: number, totalThoughts: number, nextThoughtNeeded: boolean` | read |

## Security & scope

- **Access scope:** read-only
- **Sandbox:** Purely in-memory: the tool records a chain of structured thoughts passed by the agent. No filesystem, network, or external state.
- **Gotchas:**
  - Does not persist between conversations — the chain lives only for the current MCP session.

## Quick answer

**What it does.** Provides a structured reasoning tool that guides the model through sequential thinking steps, with revision and branching support.

**Best for:**
- Multi-step problem solving
- Debugging complex bugs
- Planning and architecture decisions
- Research synthesis
- Agent task decomposition

**Not for:**
- Simple factual lookups
- Creative tasks
- Single-turn conversations

## Recipes

### Plan a migration in explicit steps

```
Use sequential-thinking to plan a 5-step migration from a REST API to GraphQL for a small CRUD service. Show each thought, note any revisions you make along the way, and finish with a single-sentence summary.
```

_Tested with: Claude Desktop, Cursor_

## Description

The Sequential Thinking MCP provides a framework for models to reason through complex problems in structured, sequential steps. It exposes a tool that forces deliberate thought chains, making it easier to audit reasoning and improve reliability on hard tasks.

## Why it matters

Complex tasks benefit from explicit reasoning structures. Sequential Thinking makes AI decisions more auditable and reliable by externalizing the reasoning process.

## Key features

- Structured reasoning steps
- Revision and branching
- Transparent thought chains
- Agent-oriented design
- Official Anthropic support

## FAQ

### Does this actually improve model accuracy?

On genuinely multi-step tasks, yes. It works by forcing the model to externalize each step as a tool call, so the reasoning is checkable and revisable instead of being buried inside a single long response. Gains are largest on debugging, multi-file refactors, and planning tasks.

### Is it useful for simple questions?

No — it is pure overhead. For lookups, one-file edits, or straight-line code generation, it burns tokens without improving the answer. Turn it off unless the prompt is genuinely hard.

### Does it support revising earlier thoughts?

Yes. The tool schema includes `revisesThought` and `branchFromThought` — the model can go back and fix a wrong step, or explore an alternative branch without abandoning the main chain.

### How does it differ from letting the model "think step by step" in plain text?

Plain-text chain-of-thought is invisible to tools and not persisted across turns. Sequential Thinking externalizes each step as a structured tool call, which means IDEs can render the chain in a collapsible panel and the model can reference earlier steps directly.

### Does it cost extra latency per response?

Yes — each "thought" is a round-trip. On hard problems that latency is well spent (fewer false starts). On easy ones it is not. Many clients let you disable it per session or gate it behind a slash command.

## Changelog

- **2026-05-31** — Refreshed install snippets and fact sheet; verified for 2026.
- **2024-11-25** — Initial directory listing.