n8

working on tools @prefectio, prev @umich —

[n8 @zzstoatzz]

@marlene_zw @nick_cloudops @fastmcp fastmcp fully supports streamable http, this seems like a deployment/config issue (e.g. reverse proxy stripping something, missing lifespan wiring if you have a parent starlette app etc, bad path)

[n8 @zzstoatzz]

@pamelafox @fastmcp logfire!

https://t.co/LLZXVqamVm

whether you meant server authors or consumers, i use it for both!

[n8 @zzstoatzz]

https://t.co/ewrjQKxGPn

iykyk

[n8 @zzstoatzz]

@Athanox_ yea it’s just fastapi dx for mcp

https://t.co/PG9R6geBVC

check out examples folder: https://t.co/cxCcJCCsMn

[n8 @zzstoatzz]

@getpy yup https://t.co/NVRiuTmACR

[n8 @zzstoatzz]

@just_cameron oh i like that

[n8 @zzstoatzz]

basically htmx but instead of a server its haiku 4.5 making it up on the fly via generateObject from @aisdk

pretty sure i saw this idea somewhere before, dont remember where tho

[n8 @zzstoatzz]

silly little gen ui thing from @cursor_ai meetup

[n8 @zzstoatzz]

@just_cameron just as a data point, excited about the memory aspect so i gave it a spin but had to stop for productivity reasons because of missing keyboard nav stuff

https://t.co/3PvVHJQnPD

[n8 @zzstoatzz]

@just_cameron *gasp* but i thought bsky was just where all the progressives ran after the election? surely i wasn't misinformed in that regard by consuming X content, surely i never would be misinformed by X

[n8 @zzstoatzz]

@tiagoefreitas @AAAzzam @zeeg @fastmcp https://t.co/kHJgxoJrg4

[David Cramer @zeeg]

I am so tired of would be AI influencers shilling GitHub as an example of why MCP is bad (and thus why CLI is better). You have all done a disservice to the community by miseducating people.

GitHub is a really poorly designed MCP. The CLI works well(-ish) because its heavily trained on. Those two factors are isolated from MCP in general.

Except people keep repeating the line "CLI is good, MCP is bad". This tells me you dont understand how agents work and why tools are needed, and why MCP is valuable

By giving an agent tools it has freedom to use those tools. They need to be well structured for a variety of reasons, but by controlling the tools you also influence _how_ and _when_ they will be used.

A CLI? The agent doesn't even know it exists. How many times have you seen them blast out to jq just to find its not installed on your system? Daily for me.

What about a CLI you just wrote? Not a chance.

Oh so you'll add `--help` and you'll shove it in your AGENTS file so it knows it exist... well it might call it.. once a while. Except it wont know how, and maybe itll call `--help` first, or maybe itll guess.

So you tell it to call `--help` every time. Great, now when RNG hits it calls and adds a bunch of context to the agent in order to know what your CLI tool does. It has to after all, unless you put all that in AGENTS.

btw, put all that in your context file and guess what you've done? Built a real shit version of MCP.

So just because GitHub - and some others - have built bad implementations on top of MCP, doesnt mean the need for tools, and the metadata required to operate said tools, is the problem.

We are still early in the lifecycle, and there's a good chance we'll be fine tuning agents on top of our tools (MCP or otherwise) to improve upon this, but any version of training is still lossy compression and will be more error prone.

So please, just stop spewing "MCP is bad" because some random dude on the internet who doesnt work for a living told you so.

[Pamela Fox @pamelafox]

I love when OSS projects make it *super easy* to copy the relevant version information for issue reports, like `fastmcp version --copy`.

[Quincy Larson @ossia]

This weekend build an MCP server for whatever project you're building right now. freeCodeCamp just published a FREE course that will show you how using the open source FastMCP Python library. 🐍

Here are this week's five freeCodeCamp resources that are worth your time 🏕️

1. freeCodeCamp just published a course on how to build your own MCP server with Python. Model Context Protocol Servers are like APIs for AI agents. Lots of developers are now building them to help agents interact with their websites' data more accurately. This course will teach you how to leverage the open source FastMCP library to build a calculator project that agents can then directly interact with. (1 hour YouTube course): https://t.co/4GIaNsTyqk

2. Learn how to pass Google's new Generative AI Leader Certification Exam. Andrew Brown is a CTO who has passed practically every DevOps exam under the sun, and he teaches this course. He'll give you a business-level understanding of Google Cloud's gen AI offerings. By the end of this course, with the help of Andrew's practice materials, you'll be ready to sit for the exam. (3 hour YouTube course): https://t.co/qnzQv7Cz2a

3. On this week's freeCodeCamp podcast I interview Evan You. He's the creator of the popular Vue front end development framework and the Vite JavaScript build tool that a lot of devs use as a boilerplate for their new projects. As a self-taught developer, he shares tips for getting involved in open source and ultimately leading open source projects and attracting sponsors. (1 hour watch or listen in your favorite podcast app): https://t.co/0cno1iRK9p

4. Teach your apps how to learn. This comprehensive Machine Learning fundamentals course will walk you through building systems smart enough to create their own algorithms. You'll use C++ to implement a Preceptron, which will then look at images of shapes and figure out ways to reliably label them. (interactive course): https://t.co/aX2qY6BN8R

5. Strix is a relatively new open source tool for testing the security of your apps and identifying vulnerabilities. It's essentially an AI-powered white hat attacker that you set loose in your codebase. This tutorial will explain how it works and how you can use it to harden your apps against common exploits. (15 minute read): https://t.co/bZDbuLsPYw

The freeCodeCamp community has avoided getting swept up in the AI hype cycle, while still finding some cool use cases for AI tools. If you enjoy our balanced, pragmatic approach to teaching these emerging tools, consider supporting our mission. Join the 10,494 kind folks who donate to our charity each month: https://t.co/PJXlqTfhI9

Quote of the Week:
“When I build things, I often start by prototyping the end developer experience I want for myself — then I reverse engineer from that to the implementation details.” — Software Engineer Evan You on this week's freeCodeCamp podcast

Until next week, happy coding.

[n8 @zzstoatzz]

@roughlyandrew terrible how?

[n8 @zzstoatzz]

@SIGKITTEN https://t.co/HtKJGkMhcg

[n8 @zzstoatzz]

can we retire the pseudo grug brain takes now

[n8 @zzstoatzz]

anti-MCP influencers be like

[Shayan @ImSh4yy]

PRO TIP: REST is overengineering. Just expose one endpoint called /api that accepts SQL queries directly.

[n8 @zzstoatzz]

@andrebuilds @TheGlobalMinima https://t.co/cf8thqW12c

[n8 @zzstoatzz]

@SIGKITTEN > this failure mode is tedious to consider as my purview

just like me fr