Top
Best
New

Posted by yarapavan 13 hours ago

You said no MCP(earendil.com)
579 points | 329 commentspage 3
jolaflow 8 hours ago|
I wish the MCP protocol supported push notifications from the app to the model. That'd make for a much nicer integration, with less overhead for this kind of interaction.
trafalgar_nah 7 hours ago|
Soon https://developers.openai.com/plugins/build/mcp-events
jolaflow 6 hours ago||
That's great!
willio58 8 hours ago||
The crazy thing with the anti-MCP wave to me could be boiled down to "you don't like it? Then don't use it."

It's like people hate it so much they forget how easy it is to spin one of these up; it's not like I spent weeks and weeks working on this.

I spun up an MCP server for a personal finance app I've been working on. OAuth with personal tokens and all the tools I added- guess how long it took me? Like 2 hours maybe. Some people find it useful, including me, and some don't. That's okay!

ricericerice 6 hours ago|
I think the concern is less over this specific feature and more about what this means for the project roadmap, now that they've chosen the route of built-in rather than first-party extensions.

Add in the context that they're also building their own inference provider[1], and there's certainly a non-zero risk that the project trends away from the barebones platform to build on that people signed up for.

[1] https://radius.earendil.com/

nialse 9 hours ago||
I'm going to pick on the example. Love both pi and use mcp, so it's a welcome change. However, I don't want to live in the world where anger is the prioritisation criteria for fixing bugs, and civil conversation about severe issues goes ignored.

    criteria: {
      none: "Neutral, factual, or friendly, even about a serious bug",
      mild: "Explicit annoyance, impatience, or disappointment",
      high: "Clearly angry, exasperated, sarcastic, or fed up",
    },
Jcampuzano2 9 hours ago|
You could turn it on its head and it could be exactly the opposite - that the people who are the most angry, sarcastic, etc are exactly the kind of issues that you should automatically close and ban from creating issues for being non-civilized productive contributors.

Doesn't have to be used for prioritization purposes. Obviously a little annoyance is to be expected at times, but you could explicitly ban egregious angry behavior in your project.

nialse 6 hours ago||
True that.
ramshanker 8 hours ago||
I am in the process of setting up a demo local-hosted-model using either vLLM or SGLang for my not-in-software-business enterprise. What harness would you recommend? I mean what are the top 5 open source ones. Trying to avoid unsustainable ( from the business model of agent developers ) harness. Personally I have used only OpenCode / Codex / ClaudeCode / Antigravity.
auspiv 7 hours ago||
also consider oh-my-pi. it is much more "batteries included" than Pi
sebastiansm7 8 hours ago||
Pi
carlsborg 13 hours ago||
This is somewhat similar to HuggingFace smolagents where the model writes code that calls tools, instead of emiting json to describe the tool call per turn. Here Codemode is one tool that the model calls when it needs to compose many tool calls, especially MCP ones. Is what i understand of this.
bloppe 4 hours ago||
> The way we like to think about MCP at this point is that it should be much closer to OpenAPI with intelligent tool discovery.

So, exactly like OpenAPI?

dingaling911 5 hours ago||
I might be dumb, but I don't quite understand why just running stuff in sub agents isn't sufficient for the context problem.
pjm331 4 hours ago|
this is how i've been solving the problem before some of these newer techniques around deferred loading etc. came out

a subagent per mcp toolset, basically sub agent per saas in the configuration at $WORK

I don't have any hard evals on this but it feels like it works more than it doesn't, especially now that subagent use has gotten better in general, but you do feel drawbacks if you ever need to do something that involves coordinating multiple sub agents, there's some communication overhead that wouldn't be there if it was just one agent with all the tools to do the whole job

melodyogonna 12 hours ago||
Good. I too I'm not a fan of MCPs, but these days I do find them useful. In Claude Code I connected to my company's MCP which made Claude Code infinitely more useful for everyday work stuff
rcarmo 12 hours ago|
Have a go at https://github.com/rcarmo/memento, I would appreciate Claude testers since I mostly use Codex. Just trying it and filing an issue about what doesn't work would be great...
samayashar 10 hours ago|
> That means tools should return structured data and tools should be discoverable by their documentation and description.

Treating MCP as a part of OpenAPI rather than a tool connector is a direction in which we're heading. It is important for the users to have the flexibility of deciding the model, work to be done and the tool call in one prompt. The framework sets up the configuration and gets the output.

imtringued 9 hours ago|
I'm pretty sure MCP will degrade to a bridge for authentication plus an adapter for the main protocol types. After that you just use whatever industry standard protocol that suits your specific application.
More comments...