17 comments

[ 2.9 ms ] story [ 41.4 ms ] thread
Sure, one agent will be fine in that setting. But the dynamics and requirements change you have multiple agents who need to coordinate tool use and task assignments.
Noob here. Can explain what you mean by CLI access? Are you talking about using CURL to interact with the Reddit API?
Users don’t understand CLI nor want to manage the systems to run CLI.

MCP provides users with an easy to use and convenient method to access data.

How do you prevent someone on Reddit from stealing all your bot's secrets by giving it instructions?
It feels like we are simultaneously massively upping th complexity on our cli / tui systems, and that the general universal utility they brought isn't quite as well and clearly cut these days.
check out clrun the CLI interface for Agents that make using TUIs super easy

commandline.run npm i clrun

I made this hopefully you agent lovers will enjoy and contribute!

The normal reaction is, there you are, I managed to do this using cli. The gratuitous ragebait way is, there you are, I managed to do this using cli, hence MCP is dead. But why?
> Even Peter Steinberger who created OpenClaw

You mean the guy who released the most unsecure tool ever? Not sure anyone should be listening to him

Bragging about abusing subreddits with spam is not my idea of an endorsement.
[dead]
I believe that this is horrible for the world.

You can disagree. But I think that this is anti-life poison.

100% ! been saying this for a while, the entire contract / protocol is usually in --help , exit codes, stdout/stderr, and decades of conventions that LLMs are already trained with;

the only place MCP wins is governance (permissions/audit/stable contracts), not capability

but for shipping fast CLIs are definitely the way to go