MCP? In 2026?
Sometimes you want a generative-AI tool to be able to access another system, but only if it needs to. (Maybe it's expensive or slow to do so.) In late 2024, Anthropic introduced Model Context Protocol for this.
People, including me, got excited and started building MCP servers. In many quarters, MCP servers are still the default solution to the "expose functionality to generative AI" problem. There are other ways to solve that problem, though:
- Skills are simple but, in my experience, very effective. (Claude Skills are a Claude thing, but other systems have analogues, and in any event you can just tell the LLM "go look in
filename.mdif you need to do X.") - Claude also has Agents, which can sometimes do what MCP servers aim to do. (Again, other systems have analogs.)
- I sometimes fine-tune a model and call into it directly in cases where others might have an off-the-shelf model call into an MCP server. Custom fine-tuning (in my case, via the OpenAI API) also feels a bit stale--I think I don't read nearly as much about people doing this--but I'm still finding it useful.
- You can just expose an API and let the tool call it the way it would call any other remote resource.
- Probably Gas Town and other orchestration tools achieve this somehow, and not with MCP.
I'm sure I'm missing some! I agree with Simon Willison:
Over time the limitations of MCP have started to emerge. The most significant is in terms of token usage: GitHub’s official MCP on its own famously consumes tens of thousands of tokens of context, and once you’ve added a few more to that there’s precious little space left for the LLM to actually do useful work.
My own interest in MCPs has waned ever since I started taking coding agents seriously. Almost everything I might achieve with an MCP can be handled by a CLI tool instead. LLMs know how to call cli-tool --help, which means you don’t have to spend many tokens describing how to use them—the model can figure it out later when it needs to.
Skills have exactly the same advantage, only now I don’t even need to implement a new CLI tool. I can drop a Markdown file in describing how to do a task instead, adding extra scripts only if they’ll help make things more reliable or efficient.
Edited 2026-01-17 to add the Gas Town example.