This feels misguided. MCP is still one of the best ways to execute deterministic sub-flows (i.e., stepwise processes) and secure tooling that an LLM would either lose itself while executing or should never access directly.
Im still struggling with understanding when MCP works better. I move everything to cli after a while. Can you give me more concrete examples? Because I don’t doubt you, I just don’t understand.
Most APIs and CLIs are not setup with clear separation of permissions, and when they have those permissions are mostly designed around human access patterns and risks, not LLM ones. The primary example of course being read-only vs write access.
MCPs have provided any easy way to side-step that baggage.
e.g. in an MCP, you have tools, those tools are usually binned into "read" vs "write". Given that, I can easily configure my tooling to give an LLM (e.g. Claude Code) unlimited read access to some system (by allowing all read-only tools) without likewise giving the LLM write/destructive access.
Obviously you can design APIs/CLIs with this in mind, but up until now that has not been a primary concern so they haven't.
That makes some sense. But one can make the argument given how easy it is to create CLI tools and add new API endpoints, enhancing them is still a better approach than creating and MCP.
I'm not pro or anti-MCP myself. I just haven't had a lot of success using them yet. I've been struggling to find the right balance and every path has lead me to a CLI tool (optionally paired with a skill).
Now I'm not using my cli tools in Claude Chat proper, but I'm not using MCPs either because they just keep failing me. This could very well be a me problem, but I'm still looking for that "ah-ha" moment.
Maybe I'm misunderstanding you, but the way you describe MCP sure sounds like it's just another RPC endpoint. Those are easy to add using traditional methods. Why deal with all the overhead of MCP for those cases?
I don't think MCPs have legs long-term, but they are a great middle ground during this somewhat turbulent transition period we find ourselves in, precisely because they are not the existing tooling (they are greenfield).
An existing company that has an API/CLI might not have everyone on the team on-board with LLMs and agents and what have you – it might be hard to get buy-in to modify those existing surface areas to be more "agent compatible".
Meanwhile, a team within a company that wants to make their company more "agent forward" can build an MCP tomorrow and it is clear what it is for: is a new surface, meant to be consumed by agents, with the needs of agents top-of-mind. It avoids confusing interop issues with existing tooling while maximizing said teams ability to ship changes/experiments quickly.
You're not wrong, but I could also argue the same about creating a new CLI or custom API. Now I know there's some expectation that APIs and CLI tools don't change willy-nilly, and that's not the case for MCP endpoints (yet), but if you clearly define the use case the outcome is the same.
But I think we generally think the same. MCP is a tax we have to pay right now to play in the whole ecosystem, but it sure doesn't feel like the right play long term.
It's the security layer that I'm most interested with MCPs. Granting full access to the CLI feels super dangerous and maybe there are options to certain commands that I want to restrict from LLM usage.
state. i'm not sure if they had that in mind but one keyword that is important here is: state. Sure CLI tools can also affect state, but in MCP service it's somewhat more natural?
