I use CLI tools for Git, Atlassian (ACLI), and Metabase, plus custom SigNoz shell scripts to drive API calls instead of an MCP server. This setup works very well at the moment for me.
Having a dedicated spec makes it easier for MCP client apps to implement some nice features out of the LLM loop (e.g. OAuth flow or elicitation ), some handy user facing features (e.g. prompts and resource), and some more advanced use-cases (like server sent notifications).
But yeah, a lot of it (including tools) maps well to HTTP semantics. In fact, at work we implement our features as HTTP APIs, and we’ve built a framework that exposes them as MCP servers.
From my limited understanding, it’s basically an API, rewritten for us by AI agents. So the humans can consume your REST API, while the agents will use the MCP to effectively do the same thing.
MCP is an API. The biggest difference is that it defines some primitives that AI clients know how to implement beforehand so every MCP server works similarly, otherwise each API implements its own way to list the tools available, change/react to changes, authorization, session management, etc. MCP *is* an API.
I'm thinking more as a provider. Why put a model on my side to handle permissions/scope/auth/actions when thats already handled (or easy enough to spin up)by a rest-inspired http api?
If you're needing to give the model direct and broad access to your SqlServer instance...I don't get it.
Dedicated customer facing API is like whitelisting approved actions; MCP feels like getting an LLM to enforce a complex blacklist.
MCP is more or less API and discovery/documentation. It’s very far from being the first technology to do that.
What it does well though is to keep things relatively simple. It is also good for things that takes time to return.
I find it competitive also for non LLM usages. I have done some PoCs with MCP instead of REST/OpenAPI, or GraphQL, or Protobuf and I liked MCP.
I use CLI tools for Git, Atlassian (ACLI), and Metabase, plus custom SigNoz shell scripts to drive API calls instead of an MCP server. This setup works very well at the moment for me.
Probably covers most use cases and is a lot cheaper and safer.
Having a dedicated spec makes it easier for MCP client apps to implement some nice features out of the LLM loop (e.g. OAuth flow or elicitation ), some handy user facing features (e.g. prompts and resource), and some more advanced use-cases (like server sent notifications).
But yeah, a lot of it (including tools) maps well to HTTP semantics. In fact, at work we implement our features as HTTP APIs, and we’ve built a framework that exposes them as MCP servers.
Whats an mcp ? Mcp connectors Iv heard of
From my limited understanding, it’s basically an API, rewritten for us by AI agents. So the humans can consume your REST API, while the agents will use the MCP to effectively do the same thing.
Is anyone forcing you to use MCP? You don't need it. If you can't see the benefit of it, don't use it.
It's why I asked the question. Maybe I missed something.
MCP is an API. The biggest difference is that it defines some primitives that AI clients know how to implement beforehand so every MCP server works similarly, otherwise each API implements its own way to list the tools available, change/react to changes, authorization, session management, etc. MCP *is* an API.
I'm thinking more as a provider. Why put a model on my side to handle permissions/scope/auth/actions when thats already handled (or easy enough to spin up)by a rest-inspired http api?
If you're needing to give the model direct and broad access to your SqlServer instance...I don't get it.
Dedicated customer facing API is like whitelisting approved actions; MCP feels like getting an LLM to enforce a complex blacklist.
Check WebMCP that ChatGPT Desktop just implemented for a good example.
Now every website can create tools that gpt can use. So it's possible to have your users use AI without the need to pay for tokens.
What’s the harm in being curious and seeing what you might be missing?
There's no harm in being curious but why frame it like the thing you don't know how to use is not useful? That's harmful.