Definitely not AI generated. I wrote this during a non-internet flight. :)

Especially portability is just not possible with Skills+CLI (yet). I can use the same MCP servers through remote MCP on my phone, web, iPad, in ChatGPT, Perplexity, Claude, Mistral and so on, which I can’t do with Skills.

Hi, author here. The “MCP has severe context bloat” problem has already been solved with tool discovery. Modern harnesses don’t load every single tool + their descriptions into the context on load, but use tool search to discover the tools lazily when they’re needed. You can further limit this by telling the LLM exactly which tool to load, the rest will stay unloaded / invisible

> But a worse problem is that the output of MCP goes straight into the context of the agent, rather than being piped somewhere else

This is semi-solved as agents and harnesses get smarter. Claude Code for example does discovery in subagents. So it spawns a sub-agent with a cheaper model that explores your codebase / environment (also through MCP) and provides a summary to the parent process. So the parent won’t get hit with the raw output log

Comments URL: https://news.ycombinator.com/item?id=46985896

Points: 1

# Comments: 0

Fresh off the press

https://configmesh.app/

Demo: https://www.youtube.com/watch?v=EeA4TTpM2wk

And a few projects I’m working on:

- https://configmesh.app utility for syncing dotfiles + application configs

- https://fixmyjapanese.com AI powered grammar correction teacher for Japanese

- https://microfn.dev Toolbox and cloud runner for tiny composable JavaScript functions

It started out as a cloud runner for small (mostly pure) composable javascript functions that we run for you in the cloud. So whenever you have some small code snippet that you want to run, you can hit up microfn.dev, paste your code and then plug it into wherever you want (pipe into the terminal, use from MCP, add a cron to it, add to Siri shortcuts, use for home automation, ping it with webhooks, etc)

Now we added agents support, so you can have autonomous AI agents take your functions and decide when to use them. Even better, you can hook microfn into a MCP and have anything that supports MCP use those functions as well!

Suppose you need a new thing that your agent should do, you could ask the agent (whether that's claude or cursor) to compose a new function, add it to microfn, then use it itself going forward!

microfn.dev wants to be a toolbox for composable small tools. Imagine a toolbox at home with a bunch of hammers and screw drivers that you collect, share, and use for different purposes.

Some actual examples:

- Pull data off services (twitter, etc)

- Store data from sensors somewhere

- Give agents tools to talk to your specific systems without needing to write an entire MCP

- Wrap complex logic (eg slack auth + sending a message to slack) into a function and add it to a Siri shortcut, so you can quickly send messages to a specific channel with a ping or curl

Still very early alpha (beta-ish), but very excited about this

It's still very heavy in development but the gist is: Say you have a cool idea for something small you want to automate or run - instead of thinking about hosting, workers, lambdas and what not, you just open microfn, open the editor (or the function generator), write your function, hit save - done! All the complexity is tucked away on microfn, and your place of use just has to authenticate with microfn and nothing else. We run it for you and keep you productive.

Now you can use that function from anywhere: From the terminal, periodically with a cron, add it to an AI agent as skill, use it through an MCP (not released yet), through Siri shortcuts, share it with your friends and so on.

Say you want to have an agent or function that gets the weather and sends it through Telegram: You can either quickly generate 2 functions through the AI function generator that get the weather ("I want a function that gets the weather for Tokyo") and another one for sending a message on telegram, or you can use what's already available (such as https://microfn.dev/david/getweathertokyo). Equipping them to an agent works like in a video game - each function is a new "skill" or "tool" the agent can use, and if someone else already has some cool skills, you can fork them without needint to re-implement everything from scratch.

So like a toolbox full of small composable hammers and tools that can be used across different scenarios and places, to be plugged into existing workflows, automation or to even be used in autonomous agents and through MCP.

Again, super heavy in development and not really a 1.0 yet, more like an early alpha, but wanted to share here anyway. Feedback greatly appreciated!

Then things like:

“Fix My Japanese” - uses LLM to correct Japanese grammar (built with Elixir LiveView): https://fixmyjapanese.com

It has different “Senseis” that are effectively different LLMs, each with slightly different style. One is Claude, one is ChatGPT.

Or a slack bot that summarizes long threads:

https://github.com/dvcrn/slack-thread-summarizer

I blog about random bits and blobs in tech. Sometimes a review, sometimes trying out something new. Wanted to try and keep it interesting and not too fixated on one category.

Last post is a mini review on a travel router, and if you need one - https://david.coffee/the-case-for-a-travel-router/

Or going through the process of getting custom molded earplugs done - https://david.coffee/my-custom-molded-attenuating-earplugs/

Or using Elixir to build a distributed ChatGPT CLI - https://david.coffee/mini-chatgpt-in-elixir-and-genserver/

Not as active as I wished

Background was that we wanted to explore providing access to ChatGPT to employees at $company, but restrict access with Google OAuth using company email, and use the company OpenAI API key on the back. So I hacked together something quick that matches our requirements and we can self-host.

It’s just amazing that non design people like me can just conjure up decent looking, and usable stuff with AI. I will definitely use DALL E much more going forward for creative work

The logos are a bit noisy and need redrawing in a proper vector tool but its a great starting point to try out different ideas immediately

(The results: https://twitter.com/dvcrn/status/1578710631838289922)

If you're into domain/service-specific email addresses, give it a go!

Like if there’s some string manipulation thing I need, instead of copy pasting some snippet, I can pull in the micro package that already does this

udm-utilities [0] basically allows you to run all kinds of stuff on it through podman.

Went from trying to teach it to mimic my ISPs router to get internet working, to banging my head against IPv6 prefix delegation, to now running adguard, homebridge and a bunch of other things across different VLANs on it.

All in all, I am happy with the result, and happier that I got dragged out of my progrmaming bubble to learn proper networking with IPv6.

[0]: https://github.com/boostchicken/udm-utilities

You have to upgrade from within the IP7 app but only if you downloaded from their site, not appstore.