A: Hey Matthias_Schwehm_-_Dein_SelbstbewusstMacher,

That's a great observation, and it highlights the two different ways services can be integrated into AgenticFlow.

First, there's the main "Connections" area in your settings. Adding your OpenAI key here creates a direct, first-class integration. This allows AgenticFlow's core platform to use OpenAI models deeply – for example, when you select GPT-4 in the standard "LLM" or "Ask ChatGPT" workflow nodes. Keys added this way will show up under the "Connected" count for OpenAI in your settings. This method benefits from potentially more optimized features like streaming or parallel calls that we build specifically for these direct connections.

Second, there are MCP (Model Context Protocol) Connections. MCP is our standardized layer for quickly adding thousands of tools, including another way to connect OpenAI (like the one you found at https://agenticflow.ai/mcp/openai). When you connect via MCP, you're essentially adding OpenAI as a specific tool that an agent can use if explicitly told to (e.g., "Use the OpenAI MCP tool to generate text..."). It will also allow using specific pre-defined actions from the MCP page in workflows once our "Run any MCP node" feature launches soon.

However, connecting via MCP does not register the key in the main "Connections" section, which is why it still shows "0 connected" there. Critically, an MCP connection cannot be used as the primary engine for agents or in the standard LLM workflow nodes. It's treated as a separate tool in the agent's toolkit, not as the agent's core brain.

So, your MCP connection for OpenAI worked immediately because you successfully added it as a tool. If you also want to use OpenAI models in the standard LLM nodes, you would need to add your API key in the main Connections -> OpenAI section as well.