Transcription: OpenAgents Episode 144 - Pylon and the Model Context Protocol

Source: https://x.com/OpenAgentsInc/status/1867458253661114610 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 144: Pylon and the Model Context Protocol We introduce Pylon... Upload date: 20241213 Transcription model: gpt-4o-mini-transcribe Generated at: 2026-06-01T15:14:08Z

Machine-generated transcript. Review speaker labels and wording before using this as quote-grade source material.

[00:00] Speaker: So you've seen our mobile app, Onyx, on the left here. And on the right here, we have our new node software for open agents. We're calling it Pylon. This is a desktop app that can run on anyone's computer, Windows, Linux, Mac. And clients can connect to it. So this has got a few things going for it. The idea is that it'll combine a few of these concepts. So we showed the NIP 90 service provider in our data vending machines video. We just kind of whipped up like a really quick TypeScript server using the Groq QAPI to do those high queues. We're going to be in this video talking about the new open protocol released by Anthropic called the Model Context Protocol. But basically, we want people to be able to run a node that can satisfy NIP 90 events on the open marketplace. It can offer unique functionality we'll describe in MCP that's kind of maybe more sensitive that you want to be able to access your own personal data living on your computer. And then a Bitcoin wallet, same seat phrase as the mobile app. This will let you run software and earn Bitcoin for doing relevant things like making API calls or running models. More on the model piece in the next video. So we have here a file system component in Onyx that is tied to the server. So the server is just my local desktop environment, but I can click through and each time I'm opening a folder here, it's making a JSON RPC call to the server saying, hey, this authenticated client is requesting this information and it just right now it's assuming that I should be able to. But this enables us to interact from a mobile app with resources on a computer. Those resources could be part of the local file system. They could be part of a database, just having access to all the things that a server possibly could provide, being able to access that on your mobile phone is quite cool. So this is building on a protocol released last month by Anthropic called the Model Context Protocol. I want to give huge props to Anthropic for releasing an open protocol that's kind of different than what a lot of the closed source AI labs have been doing to this point. Let's go through it. So today we're open sourcing the Model Context Protocol, a new standard for connecting AI assistance to the systems where data lives, including content repositories, business tools, and development environments. Its aim is to help frontier models produce better, more relevant responses. As AI assistants gain mainstream adoption, the industry has invested heavily in model capabilities, blah, blah, blah. Yet even the most sophisticated models are constrained by their isolation from data, trapped behind information silos and legacy systems. Every new data source requires its own custom implementation, making truly connected systems difficult to scale. MCP addresses this challenge and provides a universal open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol. So you can read on the website, modelcontextprotocol.io. This is very well documented. And you have the specification here. And I was able to today just like literally find somewhere here, they share the schema. And I just like pasted the schema into OpenAgents and had it like just keep iterating until it built me like a Rust server. So they offer like a Python TypeScript example servers, but I'm like, no, I want to do it in Rust. I want to do it in a desktop app using Tari and TypeScript in the front end. So this is a, to my knowledge, the very first Rust MCP server connected to what I think is the first mobile MCP client. So you can read more about the, you know, you can read more about the architecture here. There's pretty pictures, but yeah, so you've got a host in this case, this is the Onyx mobile app. And then each of those can have like different MCP clients that connect to servers. So in this example, there's a local machine. This is for connecting to files and Git. Let's say if you want to be interacting with a code base, instead of just connecting out to the GitHub API, like you can just connect it to where the files actually live for a lot faster access. In this other example, it's like, hey, you want to connect to a database, be able to do any kind of querying of any kind of data that you can imagine. And then you can imagine there's a internet connected server that connects to external APIs. Like we use the Grok API to generate that haiku. So you can envision that you could have potentially like trusted connections to servers that you whitelist and that you know about. You either run them yourself or you trust that person and connect directly to them via MCP. Or if you want to kind of throw things out to the marketplace, that's where the data vending machines come in. And there may be potentially other kind of synergies between the DBM and MCP specs that we can explore over time. But we've got some docs in here about kind of like, hey, here's what NIP89 and 90 are good at. Here's what model context is for. But this is a lot about kind of, you know, discovery. So it's interesting that like Claude, the Anthropic people, like they released with this very basic implementation that's envisioned to work, at least in the short term, only on your local environment because they're in like developer preview and stuff. So there's no real discoverability. The big idea that we had previously of like connecting all these different services into one market. Well, just now we have like additional mechanisms for client server communication that if we combine that with the discoverability that we, you know, pioneered through the NIP89 and 90 in the data vending machines video. And then pay that combine that also with payments. Now that we have a built in Bitcoin wallet, did not get a chance to connect this yet to our wallet, but I think that will be what we show off in tomorrow's video. Yeah, we'll pause it there. Lots of possibilities here. We are going to launch the Pylon desktop app approximately the same time as the mobile app so people can run this themselves and kind of get earn some sats for things like, let's say that you have an OpenAI API key and you pay a tiny amount per call. You can charge a little bit more than that per call in Bitcoin sats. And someone who wants to be able to use your key can do that. How is that compliant with their terms of service? I don't quite know, but let's give it a try. We'll see you soon.