There's no real benefit to the MCP protocol over a regular API with a published "client" a local LLM can invoke. The only downside is you'd have to pull this client prior.
I am using local "skill" as reference to an executable function, not specifically Claude Skills.
If the LLM/Agent executes tools via code in a sandbox (which is what things are moving towards), all LLM tools can be simply defined as regular functions that have the flexibility to do anything.
I seriously doubt MCP will exist in any form a few years from now
2. it was delegation to a subagent to select the tools that should be made available, which sounded like it got the whole list and did "rag" on the fly like any model would
You're going to want to provide your agent with search, rag, subagent context gathering (and pruning/compation/mgmt) that can work across the internet, code bases, large tool/skill sets, past interaction history. All of this can be presented as a single or few tools to your main agent and is the more meta-pattern/trend emerging
It's a much better system in my experience.
What they said was don't pollute your context with lots of tool defs, from MCP or not. You'll see this same problem if you have 100s of skills, with their names and descriptions chewing up tokens
Their solution is to let the agent search and discover as needed, it's a general concept around tools (mcp, func, code use, skills)