MCP Simple Openai Assistant

What andybrandt/mcp-simple-openai-assistant MCP server does

The andybrandt/mcp-simple-openai-assistant MCP server exposes OpenAI assistant management and conversation features through the Model Context Protocol. An MCP client can create an assistant with a name, instructions, and model; list assistants associated with the configured API key; retrieve assistant details; and update an assistant's name, instructions, or model.

Conversation work is organized around persistent assistant threads. The server creates named threads with descriptions, records their OpenAI IDs and metadata locally, and lets clients list or delete those managed threads. Messages can then be sent to a selected assistant inside a thread.

How it works

The usual workflow is to create a managed thread with create_new_assistant_thread, use list_threads to find an existing thread, and call ask_assistant_in_thread to continue the conversation. The response tool streams assistant output as it arrives rather than waiting for the complete response. The README describes this as a way to provide progress updates and reduce timeout problems when OpenAI assistants take longer to respond.

OpenAI does not provide thread listing through its API, so the andybrandt/mcp-simple-openai-assistant MCP server maintains a local SQLite database for thread IDs, names, descriptions, and last-used information. Deleting a managed thread removes it from both OpenAI's servers and that local database.

Setup and configuration

Install the published Python package with:

pip install mcp-simple-openai-assistant

The server requires the OPENAI_API_KEY environment variable. For Claude Desktop, the README shows launching the module with Python:

{
  "mcpServers": {
    "openai-assistant": {
      "command": "python",
      "args": ["-m", "mcp_simple_openai_assistant"],
      "env": {
        "OPENAI_API_KEY": "your-api-key-here"
      }
    }
  }
}

On Windows, the Python executable may need an explicit path. The README also notes that some Windows installations may require PYTHONPATH to point to the relevant site-packages location. An alternative installation path is the Smithery CLI, but the package's own manual installation uses Python's package installer.

Tools and capabilities

The available MCP tools are:

• create_assistant: creates an OpenAI assistant from a name, instruction set, and model.
• list_assistants: lists assistants available to the configured API key.
• retrieve_assistant: returns details for a specified assistant.
• update_assistant: changes an assistant's name, instructions, or model.
• create_new_assistant_thread: creates a named, persistent conversation thread.
• list_threads: lists locally managed threads and their metadata.
• delete_thread: deletes a thread remotely and locally.
• ask_assistant_in_thread: sends a message and streams the assistant's response.

Limitations and notes

The server requires access to an OpenAI API key, so usage depends on the OpenAI account and API configuration behind that key. The current README lists model discovery, assistant fine-tuning parameters such as temperature and top_p, reading old thread history without sending a message, and file uploads as unfinished or planned work. Those capabilities should not be assumed to be available through the current tools.

The documented client setup specifically covers Claude Desktop. The server can also be used by other MCP clients in principle, but the supplied material does not provide client-specific configuration for them.