OpenAI adds MCP support to Agents SDK

#1

Post by gronky_ » Wed, Mar 26, 2025, 6:55 PM UTC

OpenAI adds MCP support to Agents SDK
openai.github.io

Re: OpenAI adds MCP support to Agents SDK

#2

Post by gronky_ » Wed, Mar 26, 2025, 6:56 PM UTC

They’re all in. They announced they’ll add support for it in the desktop app and the API in the coming months: https://x.com/OpenAIDevs/status/1904957755829481737

Re: OpenAI adds MCP support to Agents SDK

#3

Post by esafak » Wed, Mar 26, 2025, 7:08 PM UTC

That makes it table stakes for any agent framework.

Re: OpenAI adds MCP support to Agents SDK

#4

Post by tcdent » Wed, Mar 26, 2025, 7:19 PM UTC

Big question in my mind was if OpenAI was going to formally endorse this (since it was created by Anthropic) but we have our answer.

MCP is now the industry standard for connecting LLMs to external tools.

Re: OpenAI adds MCP support to Agents SDK

#5

Post by simonw » Wed, Mar 26, 2025, 7:23 PM UTC

"Think of MCP like a USB-C port for AI applications."

Given the enormous amounts of pain I've heard are involved in actually implementing any form of USB, I think the MCP community may want to find a different analogy!

Re: OpenAI adds MCP support to Agents SDK

#6

Post by jswny » Wed, Mar 26, 2025, 7:25 PM UTC

Does anyone know how MCP servers would be used via the API?

I thought they ran locally only, so how would the OpenAI API connect to them when handing a request?

Re: OpenAI adds MCP support to Agents SDK

#7

Post by peterldowns » Wed, Mar 26, 2025, 7:32 PM UTC

jswny wrote: post #6

Does anyone know how MCP servers would be used via the API? I thought they ran locally only, so how would the OpenAI API connect to them when handing a request?

You'd use a client which runs locally to coordinate between the LLM/agent and the available tools, similarly to how it's described here https://modelcontextprotocol.io/quickstart/client

There are a variety of available clients documented here https://modelcontextprotocol.io/clients

If you haven't tried any of these yet, the first place to start is Claude Desktop. If you'd like to write your own agents, consider https://github.com/evalstate/fast-agent

EDIT: I may have misunderstood your question. If you're asking "how can I make an API call to OpenAI, and have OpenAI call an MCP server I'm running as part of generating its response to me", the answer is "you can't". You'll want a proxy API that you call which is actually an MCP client, responsible for coordinating between the MCP servers and the OpenAI API upstream agent.

Re: OpenAI adds MCP support to Agents SDK

#8

Post by larodi » Wed, Mar 26, 2025, 7:32 PM UTC

Claude is like years ahead of everyone else with tools and agentic caps.

Re: OpenAI adds MCP support to Agents SDK

#9

Post by bibryam » Wed, Mar 26, 2025, 7:34 PM UTC

jswny wrote: post #6

Does anyone know how MCP servers would be used via the API? I thought they ran locally only, so how would the OpenAI API connect to them when handing a request?

command mode is for local, and SSE for remote

Re: OpenAI adds MCP support to Agents SDK

#10

Post by jtrn » Wed, Mar 26, 2025, 7:34 PM UTC

I hoped OpenAI would support OpenAPI for connecting to tools. Having created a couple of MCP servers, it feels like a less flexible and worse documented API to me. I can’t really see anything that is made better by MCP over OpenAPI. It’s a little bit less code for a lot less options. Give it some time and it will also get Swagger built in.

It’s solving a problem that was already robustly solved. So get we go with another standard.