Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

MCP is a fad

tombedor.dev

Post reply on HN
Print view View original

121–129 of 129 posts

Re: MCP is a fad

#121

Post by vidarh »

jjfoooo4 wrote: post #111
vidarh wrote: post #3

This article could really mostly be reduced to the last two paragraphs, but then it calls skills "over-engineered". Skills are basically just having the agent read the front matter with instructions to read the rest if a given skill seems useful in a given context... I don't know how it could be more minimal.

(I'm the author) Looking at the post again, I think I agree that calling Claude Skills overengineered is too harsh. I think Skills is definitely an improvement over MCP. However I still think it's a generally a mistake to put useful commands and documentation in AI-specific files. In my opinion a better approach is to optimize the organization of docs and commands for human usability, and teach the AI how to leverage…

There's no reason they need to be treated as AI specific. They're just a description in markdown with a tiny frontmatter after all.

Re: MCP is a fad

#122

Post by bpavuk »

ndr wrote: post #22
bpavuk wrote: post #8

Earlier quoted context omitted.

strongly agree! can't see any use of LLMs beyond tabcomplete and navigating an unknown codebase.

Assuming that wasn't trolling, what's the last thing you tried and when? Latest Claude Code can do a lot over lots of domains. I recommend giving their top plan a fair chance for a month. Also most people who I see succeed from the start are technical managers. That is people who can code, who are used to delegate (good at phrasing) and are more likely to accept something that works even though is not character by ch…

> Assuming that wasn't trolling, what's the last thing you tried and when?

vibe-coding the GTK-rs app with Codex, delegating ready Figma designs to Junie to implement in Jetpack Compose, improving a11y and perf on my personal website built in Svelte with Claude. (a friend threw $100 into that slot machine and shared access to Claude Code with me.)

all in December 2025. all failed *horrendously*. Claude Sonnet 4.5 actually worsened the Lighthouse performance score on mobile from 95 down to 40!! though, it did a decent job on a11y.

Codex CLI constantly used older versions of all sorts of crates, spat out deprecated code, outright refused to use Blueprints and Context7, but some time after, I could get a mockup of Superhuman-style email client.

and Junie... well, it did best. it had Gemini 3 Flash picked as a model. despite Material 3 Expressive being beyond Gemini's knowledge cutoff, it actually used Context7 to fill the gap. though, it fumbled the bag with details, performance, and overabstractioning.

> I recommend giving their top plan a fair chance for a month.

as I said, my friend already did. we both find it the worst spend of the decade. seriously, in yesteryear's Call of Duty we would at least have fun. being a purchase worse than Call of Duty is a floor so low, you have to break it deliberately.

before you say "try again," do your research around food prices and salaries in Ukraine. (or delegate it to Claude Deep Research feature, you may love it.) to feed yourself well here, you have to spend measly by Western standards $300-600/mo. (that's without housing and utilities, presuming you live alone.) though, salaries outside dev jobs are just as measly, and then you have the overall job crisis in Ukraine overlaid on top of the worldwide IT job crisis.

4-8 months of Claude Max amount to the rumored price of Steam Machine. I would rather spend $800 to see the joy on my younger sister's face and familiarize her with true games, beyond what's on Google Play.

> That is people who can code, who are used to delegate (good at phrasing) and are more likely to accept something that works even though is not character by character what they expected.

I would rather spend that time nursing a real open-source contributor or a junior colleague. I would have a real person with real willpower and real ideas in my network, and that would spread the joy of code.

Re: MCP is a fad

#123

Post by mgaunard »

zo1 wrote: post #97
mgaunard wrote: post #67

Earlier quoted context omitted.

Good software isn't about quantity but quality of the code. AI cannot produce better quality code than someone who is actually qualified in the problem domain. What I've seen AI be very good at is creating a lot of legacy code very quickly, which itself needs extensive use of AI just to maintain it. A decent approach to move quickly for PoC or prototypes, or to enable product managers to build things without a team.…

Have you been in the same industry as the rest of us? 90% of all developers out there in the wild create "legacy code very quickly" anyways, they too create "slop" before we coined the term "AI slop". This mythical " someone who is actually qualified in the problem domain " you mention is maybe 5% of the entire software development ecosystem. If you work with only those developers, you're extremely privileged and luc…

If you work on meaningful tech projects, where tech is a real driver to the businesses and there are genuine challenges to overcome, then you can't afford slop.

I say that, but then it's true I have seen businesses be successful despite low quality of software. It turned out that for those businesses, the value wasn't that much driven by tech after all.

Re: MCP is a fad

#124

Post by the_mitsuhiko »

p337 wrote: post #117
the_mitsuhiko wrote: post #54

Earlier quoted context omitted.

> MCP allows any client (Claude, Cursor, IDEs) to dynamically discover and interact with any resource (Postgres, Slack) without custom glue code. My agent writes its own glue code so the benefit does not seem to really exist in practice. Definitely not for coding agents and increasingly less for non coding agents too. Give it a file system and bash in a sandbox and you have a capable system. Give it some skills and i…

You end up wasting tokens on implementation, debugging, execution, and parsing when you could just use the tool (tool description gets used instead). Also, once you give it this general access, it opens up essentially infinite directions for the model to go to. Repeatability and testing become very difficult in that situation. One time it may write a bash script to solve the problem. The next, it may want to use pyth…

> You end up wasting tokens on implementation, debugging, execution, and parsing when you could just use the tool (tool description gets used instead).

The token are not wasted, because I rewind to before it started building the tool. That it can build and manipulate its own tools to me is the benefit, not the downside. The internal work to manipulate the tools does not waste any context because it's a side adventure that does not affect my context.

Re: MCP is a fad

#126

Post by fergie »

karussell wrote: post #102
fergie wrote: post #64

You only have to spend 5 minutes browsing for MCP servers to see that there is an issue with AI slop. MCP is probably the first "standard" to be built out in the vibe-coding era and it really shows. As mentioned in the article, its not clear to me what the advantage over OpenAPI is. Surely a swagger file solves more or less the same issue. That said, one minor nice thing about the MCP servers is that they operate loc…

What do you mean with "locally over stdin/stdout"? This is only true if the MCP server (and original service) runs locally.

It sounds weird, but for reasons that I don't fully understand (bandwidth issues maybe?), first gen MCP servers run and accept "queries" locally and talk to their AI-minds themselves on the back end. Its not an HTTP API direct to a remote service as you might expect.

Re: MCP is a fad

#127

Post by p337 »

the_mitsuhiko wrote: post #124
p337 wrote: post #117

Earlier quoted context omitted.

You end up wasting tokens on implementation, debugging, execution, and parsing when you could just use the tool (tool description gets used instead). Also, once you give it this general access, it opens up essentially infinite directions for the model to go to. Repeatability and testing become very difficult in that situation. One time it may write a bash script to solve the problem. The next, it may want to use pyth…

> You end up wasting tokens on implementation, debugging, execution, and parsing when you could just use the tool (tool description gets used instead). The token are not wasted, because I rewind to before it started building the tool. That it can build and manipulate its own tools to me is the benefit, not the downside. The internal work to manipulate the tools does not waste any context because it's a side adventure…

Maybe I'm not understanding the scenario well. I'm imagining an autonomous agent as a sort of baseline. Are you saying the agent says "I need to write a tool", it takes a snapshot, and once it's done, it rewinds to the snapshot but this time, it has the tool it desired? That's actually a really cool idea to do autonomously!

If you mean manually, that's still interesting, but that kind of feels like the same thing to me. The idea is - don't let the agent burn context writing tools, it should just use them. Isn't that exactly what yours is doing? Instead of rewinding to a snapshot, I have a separate code base for it. As tools get more complex, it seems nice to have them well-tested with standardized input and output. Generating tools on the fly, rewinding, and using tools is just the same thing. You even would need to provide some context that says what the tool is and how to use it, which is basically what the mcp server is doing.

Re: MCP is a fad

#128

Post by the_mitsuhiko »

p337 wrote: post #127
the_mitsuhiko wrote: post #124

Earlier quoted context omitted.

> You end up wasting tokens on implementation, debugging, execution, and parsing when you could just use the tool (tool description gets used instead). The token are not wasted, because I rewind to before it started building the tool. That it can build and manipulate its own tools to me is the benefit, not the downside. The internal work to manipulate the tools does not waste any context because it's a side adventure…

Maybe I'm not understanding the scenario well. I'm imagining an autonomous agent as a sort of baseline. Are you saying the agent says "I need to write a tool", it takes a snapshot, and once it's done, it rewinds to the snapshot but this time, it has the tool it desired? That's actually a really cool idea to do autonomously! If you mean manually, that's still interesting, but that kind of feels like the same thing to…

> Are you saying the agent says "I need to write a tool", it takes a snapshot, and once it's done, it rewinds to the snapshot but this time, it has the tool it desired? That's actually a really cool idea to do autonomously!

I'm basically saying this except I currently don't give the agent a tool yet to do it automatically because it's not really RL'ed to that extend. So I use the branching and compaction functionality of my harness manually when it should do that.

> If you mean manually, that's still interesting, but that kind of feels like the same thing to me.

It's similar, but it retains the context and feels very naturally. There are many ways to skin the cat :)

Re: MCP is a fad

#129

Post by aschuth »

the_mitsuhiko wrote: post #54
kburman wrote: post #18

This analysis dismisses MCP by focusing too narrowly on local file system interactions. The real value isn't just running scripts; it's interoperability. MCP allows any client (Claude, Cursor, IDEs) to dynamically discover and interact with any resource (Postgres, Slack) without custom glue code. Comparing it to local scripts is like calling USB a fad because parallel ports worked for printers. The power is standardi…

> MCP allows any client (Claude, Cursor, IDEs) to dynamically discover and interact with any resource (Postgres, Slack) without custom glue code. My agent writes its own glue code so the benefit does not seem to really exist in practice. Definitely not for coding agents and increasingly less for non coding agents too. Give it a file system and bash in a sandbox and you have a capable system. Give it some skills and i…

The skill/CLI argument misses what MCP enables for interactive workflows. Sure, Claude can shell out to psql. But MCP lets you build approval gates, audit logs, and multi-step transactions that pause for human input.

Claude Code's --permission-prompt-tool flag is a good example. You point it at an MCP server, and every permission request goes through that server instead of a local prompt. The server can do whatever: post to Slack, require 2FA, log to an audit trail. Instead of "allow all DB writes" or "deny all," the agent requests approval for each mutation with context about what it's trying to do.

MCP is overkill for "read a file" but valuable when you need the agent to ask permission, report progress, or hand off to another system mid-task.

Post reply on HN