[Music] Cool. Thank you everyone for coming. Um, my name is Chris and I'm from OpenC and I work on AI products uh with our team and I've got some cool things to share with you today. um primarily OpenC MCP a solution we've been building for AI agents to easily access and work with our data. Um I have to show this video contractually.
Um it's a very good video. Um but yeah um if you want to follow me after this talk you can find me on Twitter and social atonan. Um but yeah, we over the past two years we've been rebuilding open C from the ground up. We've launched OS2 to unify not just NFTts but also tokens and DeFi assets. Today we support over 20 blockchains.
We have NFTTS, tokens, DeFi. Soon we will have advanced features for token trading. We will have AI agents and we have an exciting new mobile app that will put web 3 in your pocket. But the main piece I'm excited about is AI. We've been thinking about this a lot internally.
And you know, if you've been following the trends, it's starting to feel like science fiction. I'm talking to my computer all day, generating code, text, images, video, and automating tasks and workflows. Our team has embraced this both in our software engineering where a growing amount of our code is AI generated um but also beyond when it comes to reviewing collections for trust and safety issues for automating creation of documentation we are leveraging AI and it's accelerating us significantly and we've also been thinking about the developer story here And this is why we have open CMCP. We want to think about how we can expose the OpenC ecosystem to this growing world of agents and MCP is our first um step in that direction. We've been building our own agents and the MCP has been critical to opening up our data and it's been so useful that we've decided to share it with the community to see what folks can build.
If you're unfamiliar with MCP, there's a few ways we can describe it. Um, at a very simple level, you can think of it as USBC for tools and applications. Um, going a level down, it's essentially an API that AI agents can use to connect to external applications. Um, you can expose tools and capabilities to your LLM and enable it to take action. Um to give you an example of how we've been working internally, you know, if we sit down to build a product, we will spend time brainstorming with the team, we'll meet, we'll record those meetings, we'll transcribe them, we'll use tools like Miro and Figma to create records and then what we typically do is throw all those in an LLM and they will consolidate that information, tie it all together and produce a PRD that is then put it in notion um via an MCP server.
We can then review that as humans to make sure that it hasn't just hallucinated a load of features and then we can start doing more traditional project management breaking down into tasks thinking about dependencies prioritization and then with another prompt we can take that and we can put all that information into linear for our human and AI developers to start building those features. And none of this would be possible without MCP. It is the glue that makes LLMs and AI agents useful and allows them to take action in this sort of software ecosystem that we live in. We also use it for things like incidents where we use data dogm to pull in information about what is failing logs from our website so that we can quickly diagnose and solve user issues. um it's changing the world and you know I think it's going to be the future as we u move to this sort of AI connected future.
So what is open C MCP? It gives your LLM's access to open C data and in some cases actions if you're familiar with that API um it is very similar. You can access collections, items, tokens, wallets. You can get analytics of onchain events and you can build everything around that. And also you can do swaps.
We can give you a transaction that you can sign and submit on chain and automate the trading and really sort of think about how agents can help you make money because we all want to make money, right? We also have another feature which is natural language queries that is in beta right now but it's available um today um where you can actually ask your LLM hey how are crypto punks doing today and it will figure out the right combination of tools and underlying API calls to make to fetch that data from open C. Um and this is just the start and as we go along we'd love to get your feedback on what is working, what isn't. If you have feedback this weekend, come by our booth. You'll get a free hat.
Um, and let's give you some demos. So, this is an app called Chorus, um, from the team at Melty Labs. Think of it like chat GPT. Um, one cool thing is it lets you talk to multiple LLMs at the same time. So, you can send a query to Claude, GPT5, Gemini, and you can see the results.
It's great for testing MCP tools and it's also what we tend to use is as a control panel for our work. Um, so to start using OpenC MCP, you can go in there, you can add the MCP URL and uh then you can get started. So let's ask it a question. So I've got Claude Opus selected. I've got a tool enabled.
That's my MCP server. And I can just go in and type. So, you know, what is the floor price of moonbirds? So, hit return. It's thinking for a bit.
It's working out which tools to call on the MCP and it's searching for a collection right now and boom, it's returned information. I don't need to think about what API calls I'm making or sort of how to get that data. I tell the LLM, it analyzes the available tools and it figures it out. And the great thing is all of our tools are self-documenting. I don't have a fancy prompt in the background which says, "Hey, if someone asks you about a collection, you need to do this."
All of our tools ship with full descriptions of the tools themselves, the parameters so that the LM can easily navigate. Let's try another example. This one is maybe a bit more dataentric. Um, so I'm gonna ask it, give me a brief analysis of my wallet. So, you know, normally you'd go to Ether Scan, you'd sort of spend a bit of time digging around, you'd probably go to OpenC, you'd go to other spots to pull that information, and in the space of about 5 seconds, I've got a full summary of my wallet, and I can continue this conversation.
I can ask it for further recommendations based on my NFT holdings. What collections might be interesting for me? What action should I be taking on my portfolio? Um, if I'm doing security analysis, I can say, you know, does this look sketchy? And it will tell me and we think that this is going to be really powerful for agents um for multiple tasks around web 3.
We can do a similar thing in cursor um or sort of clawed code. So you would just add a JSON config pointing to the MCP server and you're able to use it. And let's sort of take a look at a quick code example. So this is using Vel's AI SDK. Um this is of a favorite of ours.
You can also do this straight with OpenAI's SDK, Claude, Cloudflare, pretty much any sort of AI agent platform supports MCP tools. So you can see we're pulling in um the right the imports at the top. We're defining our client. Um here we're using SSE um no particular reason. Our MCP server supports both HTTP streaming and SSE connections.
And then we're um grabbing the tools. And here is where you can do sort of cool things like filtering the tools if you don't want to expose certain things to your LLM. That's often useful if you don't want to sort of have the possibility that it goes down a rabbit hole and uses the wrong tool. But generally, we found the newer models like GPT, Opus, and Sonnet are very good with tool calling. And then I'm doing my um agent query here.
Um so I'm using GBT5. I've got a user message. What are the trending collections right now? And I'm giving it the tools. If I want to, I can um require tool usage.
that's sometimes useful if you find the LLM isn't using the MCP tools um and it's trying to answer from its own knowledge but again most of the time with the newer models that's not necessary and then I'm also setting maximum number of tools as 20 and that means that you know it won't get carried away with tool calls it'll stop after a predefined amount and then I can just run my um call and get a response back and it's as simple as that to integrate with an MCP server. You can integrate with multiple servers. So if you're orchestrating a workflow between OpenC, maybe Alchemy for blockchain analytics, maybe you're putting all the data back into Notion, you can add all of those and the LLM will automate that workflow um for you. And that is um just really smooth and really cool. In terms of what we have available, we're focusing right now on five broad categories.
First is market insights. So you can get what is happening right now and you can also control that somewhat. So you can get the top collections of the past hour, day, week, month. Same with tokens. You can get things that are trending and those use the similar ranking that we use on openc.
io um for our consumer product. And that really sort of is great if you're building an agent that um is monitoring latest trends and taking action from that. We obviously support search and discovery. Um we have tokens, we have items, we have collections, and then that sort of free text AI enhanced search that I was telling you about um before. If you know what you're looking for, and this is typically by contract address or open C slug, you can look up specific entities.
So like the example what's the floor price of moonbirds and with all these requests you have additional options so that you can fetch activity of a wallet holders listings metadata traits. If you want to build something that looks at the rare traits and trade floor prices of a collection you can do that. So I can see okay how much is my diamond skeleton moonbird selling for today? Um, and it will give me that answer. Um, it is very good at feriting around and finding data.
Earlier today, um, someone asked how what was the latest sale price for a founder CryptoKitty. Um, I don't know if you're familiar with the collection. There's a hundred of those. Our LLM told us within a minute that actually there haven't been any sales recently. So, the last sale was at a certain price, but that's probably not valid anymore.
So it can handle all sorts of queries. If you're more interested in individual portfolios and wallets, then we can also get the NFT token balances for their for those and then you can trigger further lookups to get more information on those assets and really build a profile of a user and a wallet. Like I say, really good for portfolio advice, really good for security analysis and lots of other use cases. And then on the trading side, we support swaps. And this is across all the 20 chains that we support on the OpenC website.
So you can get a transaction, you can execute a swap, you can make instant decisions powered by AI instead of the human flow where, you know, you're never going to be as fast as the bots, but this gives you at least a fighting chance. And you can automate things like stop-loss and um certain of actions you want to take in your portfolio. In terms of use cases, our team has brainstormed a lot of these and I I think the list goes on. If you are really interested in NFTTS, you can look at sniping. You can look at watching um popularity of collections over time.
Um you can look at fetching data from a single collection, maybe feeding it to a video or image generator and then you have a talking moonbird that you can then share on social media. If you're doing token stuff, you can check for rugs. You can sweep your dust. You can rotate your portfolio. You can buy the dip.
If you're more interested in users, this helps you find super fans. It helps you list all the holders who have a certain collection or a certain trait in a collection. You can fetch that data and start thinking about what you can do with it. And you can have a truly personalized discovery experience. Um, and you know, part of the impetus of releasing this here today is to see what people do with it.
Uh, we're really excited with the use cases that folks, uh, come up with. With that said, we have a $10,000 prize. There's a first place of six grand, then three, then two for the best AI experiences people make with Open C MCP. Please don't make us take all that money back to the office. we just don't know what to do with it.
Um, we really hope we can give it to some We really hope we can give it to some teams who make awesome agents that blow us away. Um, and yeah, that's um it from me. If you want to see our documentation, scan that QR code. There's a um link where you can get an access token and beta access to OpenCMCP. We'll be opening this up to a wider audience over the next few weeks.
And then stay tuned for more exciting agent products from OpenC. Any questions?
Of all the projects you've seen or you've experimented with, which one personally excites you the most?
Let's see what has excited me the most. Um, I don't want to give away anyone's ideas. Um, I like the ones that combine multiple steps in the workflow. It's like if we come out of here and I see sort of a few dozen chat bots, I'll be disappointed. If I see talking moonbirds fed by real-time data, I'll have a smile on my face.
Um, if I see agents that are automating trading, I think that can be interesting. I think you need to think about like how that sort of differs from like Bankerbot and some of the other things out there. Um, but no, I we don't know what the community is going to build and I think that's part of the thrill of it. Um, we want to see something new. We want to see something cool that combines many aspects of AI and web 3.
So, how how do we sorry, how do we get the access?
Yeah, there's a yellow box at the top and there's a slightly hard to read link in there. If you tap that um you'll get a token um very quickly. [Music]
Automatic transcript — names and jargon may be misspelled.