A colleague suggests avoiding scope collisions by giving the same underlying server a different name at each scope. What does that actually change?

The principle — What counts as a duplicate is a naming question, so renaming is not a workaround. It changes the answer.

Why the key is correct — The three scopes match duplicates by name. Two different names are therefore two different servers, and both load, each with its own tools listed under its own label. That may be what your colleague wants, but it is the opposite of avoiding a collision — it guarantees you connect to the endpoint twice.

Why the others are wrong — Matching across the three scopes is by name rather than by URL. There is no numeric-suffix disambiguation across scopes. And a collision does not stop the session; it is resolved quietly by precedence.

Remember this — Renaming does not dodge a collision so much as remove it, and what you get instead is two connections to one endpoint.

Sources — Anthropic's Claude Code MCP reference.