What I was doing: Creating a new MCP server connection to connect a third-party MCP server (Stairoids) to a scenario, via Credentials → Credential requests → [request] → MCP Client “Call a tool” module.
What is the problem & what have you tried?
Result: Toast error “Validation failed for 3 parameter(s). Bad Request”
All three parameter names sent by the frontend (MCP server, serverUrl, accessToken) are rejected by the backend as “Unexpected parameter” — the backend’s validation schema doesn’t recognize any of them. This points to a mismatch between the frontend’s request format and the backend’s expected schema, not a user input error. It’s reproducible regardless of which URL or token value is entered.
The problem is not your URL or token. Make.com’s front end is sending 3 pieces of information that its backend doesn’t understand. It’s like giving the right form to a computer, but using labels the computer doesn’t recognize. The API fields need to be fixed or mapped correctly.
The toast is on POST /api/v2/credential-requests/credentials/{requestId}. That is completing an existing request from the MCP Client form, not Create new Credential Request. Those are two different calls.
If the form still posts MCP server, serverUrl, and accessToken, and the backend calls all three unexpected, no URL or token value will pass.
Has anyone had this connection actually save after that toast, or only after sending the same payload with the keys the schema currently lists?
That’s incredibly frustrating when you’re just trying to get a connection set up and the platform itself is throwing a validation error. I’ve run into similar issues with API integrations before, where the frontend and backend just aren’t speaking the same language. It completely kills your momentum. The fact that the frontend is sending “MCP server” while the backend is expecting something else is a classic case of a naming mismatch. It means you’re stuck waiting for Make’s dev team to align their own internal schemas, and there’s not much you can tweak on your end to fix it. This whole situation reminds me of setting up a solid conversational AI system. The tech is only as good as the data it’s given. If the inputs are mislabeled or formatted incorrectly, the whole conversation breaks down. You spend more time debugging the tools than actually using them to connect with your audience. Hopefully, they patch this up quickly because it’s a real roadblock.