Feature Request: Native Support for MCP Server Integration in OpenAI Python SDK

[iceriny]

Confirm this is a feature request for the Python library and not the underlying OpenAI API.

• This is a feature request for the Python library

Describe the feature or improvement you're requesting

Description
I propose adding native support to the OpenAI Python SDK for interacting with local or remote Model Context Protocol (MCP) servers. This integration would enable OpenAI SDK users to seamlessly leverage MCP-provided resources (prompts, tools, models, etc.) while retaining compatibility with OpenAI’s existing interfaces.

Proposed Functionality

1. MCP Server Configuration: Allow users to configure an MCP server endpoint or local process (via parameters like StdioServerParameters in the example) directly within the OpenAI client initialization.
2. Protocol Bridging: Automatically translate OpenAI SDK requests (e.g., chat.completions.create) into MCP protocol calls (e.g., prompt resolution, resource loading, tool execution) when an MCP server is configured.
3. Backend Agnosticism: Retain the default OpenAI API behavior unless an MCP server is explicitly specified, ensuring backward compatibility.

Use Case Example
A developer could initialize the OpenAI client with either:

# Default OpenAI behavior  
client = openai.OpenAI(api_key="...")  

# MCP-integrated behavior  
client = openai.OpenAI(  
    mcp_server=StdioServerParameters(command="python", args=["mcp_server.py"])  
)  

Subsequent client calls would then utilize MCP-managed resources (prompts, tools), enabling dynamic model orchestration while maintaining familiar OpenAI SDK semantics.

Benefits

1. Simplifies integration workflows for users operating in hybrid OpenAI/MCP environments.
2. Enables use cases like runtime prompt templating, resource management, and tool execution via MCP.
3. Strengthens interoperability between OpenAI’s ecosystem and the MCP protocol’s extensibility.

References

• MCP Python SDK: https://github.com/modelcontextprotocol/python-sdk

This feature would empower developers to flexibly combine OpenAI’s capabilities with MCP-managed infrastructure while maintaining a unified coding interface.

Additional context

No response

[RobertCraigie]

Hey @iceriny, thanks for the request.

At this time we're not planning on building in any support for MCP in this SDK. Instead that belongs in a higher-level framework like the Agents SDK https://github.com/openai/openai-agents-python. I'd recommend checking that out.

[jlvanhulst]

I am a little surprised to hear that! IMO it would make things so much easier (especially as MCP matures, obviously) around Responses API as well? Tool call handling through MCP seems such a natural solution that ALSO makes it 'trivial' to do better/easier testing in the playground? Just enable the (URL based obviously, not local) server and you're good test.

[ycjcl868]

Response API already supports MCP, https://openai.com/index/new-tools-and-features-in-the-responses-api/

response = client.responses.create(
  model="gpt-4.1",
  tools=[{
    "type": "mcp",
    "server_label": "shopify",
    "server_url": "https://pitchskin.com/api/mcp",
  }],
  input="Add the Blemish Toner Pads to my cart"
)

[jlvanhulst]

Yes since yesterday! I am very excited about that!