Local stdio
Sequential Thinking
Reference MCP server that gives the model a numbered, revisable chain of thought for problems that should not be one-shot.
Published as @modelcontextprotocol/server-sequential-thinking. The model calls a single tool with a thought number, a running total, and optional revision or branch fields. History lasts for the connection and is capped by MAX_THOUGHT_HISTORY (default 1000).
The package asks for Node.js 24 or later when you launch it with npx. Set DISABLE_THOUGHT_LOGGING=true if you do not want each thought printed to stderr.
Setup snippet
{
"mcpServers": {
"sequential-thinking": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
}
}
}Client config shape varies. Cursor and Claude Desktop use mcpServers. VS Code often uses servers. The values above are the part that matters.
- Homepage
- modelcontextprotocol.io/
- Maintainer
- MCP steering group
- Transport
- Local stdio
- Auth
- No token. The thought history stays in the server process for the life of the connection.
- Last verified
- Tags
- Reference, Local, Reasoning