Does MCP server call any external OpenAI API?

[kapricanon]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

I want to develop an application which consist of a Interface like Claud Desktop, MCP Server like OpenAPI which can connect to all my API yaml in my organisation's GitHub & the infrastructure where the APIs are running so users can query or execute the APIs by giving commands in Claud Desktop.

My question is, if I install all these components within my organisation's infrastructure, is there any point where MCP or any other component needs to call any external API like OpenAI API? I am more concerned from a security perspective and whether any internal data/code will be sent externally?

[olaservo]

MCP is a communication protocol that just defines how different components should talk to each other, so its really up to the implementation whether a given component would call an API, or follow the security guidelines defined in the MCP spec. So while MCP provides guidance on security boundaries, the actual data flow is determined by the components you implement or install, and how they're configured to operate.

If the MCP servers you're using need to use an LLM and they use Sampling as defined in the MCP spec, then the client component (your Claude Desktop equivalent) would be responsible for handling the LLM calls. Otherwise, its also possible that a server could make calls to an external LLM (like OpenAI API). But in the end what happens with your data is still up to the server/client implementation and/or the servers you choose to install within your organization's infrastructure.