MCP Server for OpenAI
This project provides a server compliant with the Machine-to-Machine Communications Protocol (MCP) that acts as a bridge to various OpenAI API functionalities. It allows MCP clients to interact with OpenAI's Chat Completion and Assistants APIs through a standardized set of tools.
Features
The server exposes several tools to interact with the OpenAI API:
- Chat Completion: Ask a direct question to a specified model (
gpt-4,gpt-3.5-turbo). - Assistant Management: Create, list, retrieve, update, and delete assistants.
- File Management: Upload, list, and delete files associated with assistants.
- Tool Management: Enable or disable tools for assistants, such as
file_search.
Installation
To install the necessary dependencies, navigate to the project root and run:
pip install .
This will install all the packages defined in pyproject.toml, including mcp, openai, and click.
Usage
To start the server, you need to provide your OpenAI API key. You can do this by setting an environment variable or by passing it as a command-line argument.
Using an environment variable:
export OPENAI_API_KEY='your-api-key-here'
mcp-server-openai
Using a command-line argument:
mcp-server-openai --openai-api-key 'your-api-key-here'
The server will start and listen for MCP messages over stdio.
Usage in Cursor
To configure this server in an MCP client like Cursor, use the following configuration. Replace "LOCAL PATH" with the absolute path to this project's directory and "OPENAI API KEY" with your actual key.
{
"mcpServers": {
"openai-server": {
"command": "mcp-server-openai",
"args": [],
"env": {
"PYTHONPATH": "LOCAL PATH",
"OPENAI_API_KEY": "OPENAI API KEY"
}
}
}
}
Available Tools
Here is a detailed list of the tools exposed by the server:
| Tool | Description | Parameters |
|---|---|---|
ask-openai | Ask a direct question. | query (string), model (enum: gpt-4.1, gpt-4.1-mini, gpt-4o, etc.), temperature (number), max_tokens (integer) |
list-assistants | List all available assistants. | None |
retrieve-assistant | Retrieve an assistant by its ID. | assistant_id (string) |
create-assistant | Create a new assistant. | name (string), instructions (string), model (string), temperature (number), file_ids (array of strings), enable_file_search (boolean) |
update-assistant | Update an existing assistant. | assistant_id (string), name (string, optional), instructions (string, optional), model (string, optional), temperature (number, optional), file_ids (array of strings, optional), enable_file_search (boolean, optional) |
delete-assistant | Delete an assistant by its ID. | assistant_id (string) |
upload-file | Upload a file for use with assistants. | file_path (string) |
list-files | List all files available for assistants. | None |
delete-file | Delete a file by its ID. | file_id (string) |
Model Pricing
Below is an estimated pricing table for some of the models available through this server. Prices are per 1 million tokens. Please verify the latest prices on the official OpenAI pricing page, as they can change.
| Model | Input Price / 1M tokens | Output Price / 1M tokens |
|---|---|---|
gpt-4o | $5.00 | $15.00 |
gpt-4o-mini | $0.15 | $0.60 |
gpt-4-turbo | $10.00 | $30.00 |
gpt-3.5-turbo | $0.50 | $1.50 |
Development
To contribute to this project, clone the repository and install it in editable mode:
git clone https://github.com/snilld-ai/openai-assistant-mcp
cd openai-assistant-mcp
pip install -e .
Testing
The project includes a basic test file to verify the connection to the OpenAI API. To run the tests, use pytest:
pytest
Make sure you have your OPENAI_API_KEY environment variable set.
License
This project is licensed under the MIT License. See the LICENSE file for details.