Sequential Thinking

Sequential Thinking MCP provides a server that enables an AI agent to perform advanced meta-cognition and dynamic, reflective problem-solving. It guides the agent through complex tasks, logs thoughts and plans, and suggests tools to execute actions, helping you build more capable autonomous workflows.

How to use

You interact with the server through an MCP client. Connect to the MCP server using one of the supported transports, then issue tasks and receive structured guidance that includes thoughts, plans, and tool recommendations. The server is designed to support iterative planning, self-evaluation, and proactive task management so your AI agent can tackle multi-step problems with clarity and accountability.

Key capabilities you can leverage include think for logging thoughts and planning, left_to_be_done for explicit future-state management, and tool_recommendation to identify which tools to use to execute planned actions. When you run tasks, the agent can refine its plan after each step and adjust to new information or errors, promoting more reliable outcomes.

How to install

Prerequisites you need before installation:

Python 3.10 or newer

A local development runtime such as uv for running and testing the server

Choose one installation method below and follow the steps precisely.

Install from PyPI (Recommended)

pip install sequential-thinking-mcp
`

Run the MCP server (default transport is stdio)

python -m sequential_thinking

For local development with uvx

uv sync --locked

uv run -m sequential_thinking

Docker deployment

docker compose up --build -d

Additional run options include using a container with a specific port mapping and transport when you need to expose the MCP server externally.

Configuration and deployment notes

You can connect to the MCP server using different transports. The following examples show how to configure connections from an MCP client.

{
  "mcpServers": {
    ...
    "sequential-thinking-mcp": {
      "command": "uvx",
      "args": [ "sequential-thinking-mcp" ]
    },
    "sequential-thinking-mcp": {
      "serverUrl": "http://127.0.0.1:8000/sse"
    },
    "sequential-thinking-mcp": {
      "serverUrl": "http://127.0.0.1:8000/mcp"
    },
    ...
  }
}

If you prefer a direct HTTP transport, you can point your client to the SSE endpoint or the streamable HTTP endpoint provided by the server. Use the SSE URL for bistreaming interactions and the /mcp URL for HTTP-based interactions where supported by your client.

Example MCP connections

{
  "type": "http",
  "name": "sequential_http",
  "url": "http://127.0.0.1:8000/sse",
  "args": []
}

{
  "type": "http",
  "name": "sequential_http_mcp",
  "url": "http://127.0.0.1:8000/mcp",
  "args": []
}

{
  "type": "stdio",
  "name": "sequential_stdio",
  "command": "python",
  "args": ["-m", "sequential_thinking"]
}

Available tools

think

Log a thought, plan next steps, and recommend tools to achieve the goal.

left_to_be_done

Manage explicit future state and task estimation to keep track of what remains.

tool_recommendation

Suggest specific tools to execute planned actions or gather necessary information.