They are working on adding some custom/community nodes to n8n cloud (See recent response on this feature request), but generally they are only allowed on self-hosted.
Not sure if this is what you are asking on the 2nd part, but you could wrap pretty much anything in a service, host it yourself, make it reachable from the public internet, and call it from n8n cloud via HTTP request node.
In the case of n8n cloud, suppose if we want to invoke an external application (similar to salesforce), what are the options available apart from introducing a custom node ?
You can do almost anything with an HTTP Request node as long as the API you need is available via REST (or a similar http-based endpoint) (even this full list for Salesforce, if you weren’t finding everything you want/need in the Salesforce node).
Suppose if we use multiple HTTP Request node for invoking APIs of different applications, then how the AI Tool Agent identify which HTTP Request node to be processed and what purpose. Does the AI Tool Agent identify the proper HTTP Request tool based on the description given in the HTTP Request tool. Or should we provide or register those information some where else in the AI Tool Agent, like the purpose of each HTTP Request tool used in the workflow ?