Sequential Thinking MCP Server (Azure OpenAI Adaptation)

This repository contains a modified version of the Sequential Thinking server for the Model Context Protocol (MCP). The primary goal of this adaptation is to make the server compatible with environments like Microsoft Azure OpenAI, which may have stricter limits on the length of tool descriptions provided to the Language Model (LLM).

Background & Origin

This project is derived from the official MCP examples provided by Model Context Protocol.

Original Repository: modelcontextprotocol/servers

Due to the large size of the original repository containing multiple examples, the relevant source code for the Sequential Thinking server was manually extracted and included here rather than creating a full fork. This allows for a focused project containing only the necessary files for this specific server adaptation.

Full credit for the original Sequential Thinking MCP server implementation goes to the maintainers of the original MCP examples repository.

Modifications for Azure OpenAI Compatibility

The standard tool description for the Sequential Thinking server can exceed the character limits imposed by some LLM APIs, such as Azure OpenAI limit for tool descriptions is 1024 characters. To work around this:

1. The main sequentialthinking tool now provides a short description.
2. A new, separate tool called get_sequentialthinking_instructions has been introduced that delivers the full instructions.
3. The description of the main sequentialthinking tool explicitly instructs the LLM agent to call get_sequentialthinking_instructions first to retrieve the full, detailed usage instructions.
4. The client interacting with the LLM and this MCP server must be capable of:
   • Recognizing the need to call the instructions tool.
   • Executing the get_sequentialthinking_instructions tool call.
   • Receiving the instruction text in the response.
   • Including the received instructions in the context for the subsequent LLM turn where the actual sequentialthinking tool is invoked.

This "instructions-as-a-tool" approach ensures the LLM receives the necessary detailed guidance while keeping the initial tool description within API limits, provided the intermediary client handles the two-step process correctly.

Installation

Add this to your MCP config json file

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

License

This adapted version retains the original MIT License. See the LICENSE file for details. (Note: You may need to add a LICENSE file if one doesn't exist).