Hey great question! It has a universal tool called plan and run-code, so basically when it gets a task it will determine what API access it has in the GitHub codespace, for example:
Write me a 10 Slides google slides presentation on the current news -> Claude will hit the plan tool -> come up with plan and recognize it has access to a codespace with the googleapis npm installed and it has a scoped Google User Token -> it will then write code to create the slide deck -> You can then approve or deny that code -> if approved the code is executed to create the slide deck -> then you can approve whether to share the result with Claude or not
Write me a 10 Slides google slides presentation on the current news -> Claude will hit the plan tool -> come up with plan and recognize it has access to a codespace with the googleapis npm installed and it has a scoped Google User Token -> it will then write code to create the slide deck -> You can then approve or deny that code -> if approved the code is executed to create the slide deck -> then you can approve whether to share the result with Claude or not