Alchemy Is Giving AI Agents Wallets, MCPs and CLIs to Act Onchain

You're watching Incline by StableAsh live, live series covering the climb toward Agentic Commerce. We got Blake Duncan, lead agent experience engineer at Alchemy. Blake, welcome to StableAsh Live. Yeah. Thanks for having me. Excited to be here. Appreciate your patience as we brought you on. I know that last one went a little long, but bring us into Alchemy, what what what the latest is. What and what even lead agent experience engineer.

Yeah. This is first time I've heard heard a role like that. What is that? I understand. What are doing? Remind me of, like, the go to market engineers that were, like, the clay masterminds. So this is, like, maybe the new wave. So we were we were kind of yapping about that earlier, trying to figure, like, what does that actually mean?

Yeah. That's actually a really good question. It's so we have at Alchemy, we have a team called the agent experience team. It's a team that we started, I think, probably back in February now.

Historically, we've had an engineering team at the company called developer experience, and it would do a lot of, like, the products that you'd expect from the developer experience team, which is, like, our dashboard, our, like, dev tools, working on just making sure that developer has everything, like, from docs to just any experience that you would have through, like, a a traditional interface with the company. And so that makes it easy for developers to use our products.

With the growth of agents, we thought we should also have a similar team for agents that want to use Alchemy. So we spun up the agent experience team. And the core focus of this team has been our MCP, which gives you access to all of our products through the MCP. It also allows you to, like, make edits to your account, change configurations, debug things. You could actually use RPCs. You could use our data APIs. You could, like, look up analytics through our MCP. We have a similar functionality.

We also built out a CLI. Those are really, the two interfaces that you if you're using TALDR codecs that you would interface with the company's products. The one thing that we additionally added to the CLI, which is really exciting, is the agent wallet. So if you download the Alchemy Alchemy CLI, excuse me, you actually can, set up a a session with a wallet that we spin up in the dashboard. And that session through your terminal has full access to your wallet capabilities.

So you can give a wallet to, Claude CLI or Codec CLI and have it take action on chain. We built, most of the things that you would need for an agent to do really anything on chain. You could swap natively within the CLI. You could bridge within the CLI. So that is definitely, like, our exciting kind of, like, experimental product that we're we're we're really seeing how people try to use this with agents. But that kind of is like the manifestation of the agent experience team.

Like, would you and this is something I use CLIs, I use MCPs, I don't really know the difference, and I don't really know when I should start to shift one way to for another. That's one question. And I think the other question I would even ask is how much of the focus right now is purely what agents can do versus what humans can do augmented with agents?

Because I think I start to go down the route of, like, these autonomous agents that all of a sudden, you know, go find the wallet and go find the tool to get the job done, whereas I feel like right now, personally, I'm in the stage of I'm telling my agents to start using these tools to then figure out what I need to get done. Does that make sense? The kind of the different approach there when it comes to, like, discovery and even just, like, how do my agents start using tools?

Do I have to give them the tool, or do they go find the tool?

Yeah. I think that's a a really good question or two questions. So the first one is CLI versus MCP. It's it's pretty funny because I feel like the the answer to this question changes, like, every month. I think originally, everyone was like, oh, you need to use MCPs, but then people were worried about, oh, the MCPs bloating your context window. But now people aren't as worried about context windows. So it's gone back and forth from MCP to CLI.

I think one trend that I have noticed that a lot of companies are adopting is the MCP is really turning into, like, the read path for everything that your company offers, where if you wanna actually take right actions, you would use the CLI. I think that's like a trend that I've kind of seen over the past month. But it is evolving. I don't like, there's not really a definitive answer as to what you should use. So it's really just kinda like use what the company has built.

And for us, our objective has been to basically just cover all the surface areas because we don't know a year from now like, maybe a year from now, everyone's using CLIs. Maybe no one's using CLIs since there's some new way to interface. Everybody's just using APIs directly and you don't even need an MCP. So, yeah, first question. Yeah. There's there's not really a definitive answer.

I it does seem like it's trending a little bit more towards like MCP's are a little bit more read heavy, CLI is a little bit more write heavy, like, edit That's good framework, I think, for me for me to start with. Like, that really simple really simple box. Yeah. Exact yeah. Mean, like, there's also different, like, release patterns because, like, for an MCP, it is kind of like an API, so you could easily update MCPs and provide new tools just from, like, a deployment to an AI agent.

So from, like, a read environments, that makes a lot of sense. Like, you just offer new data to to agents. But for, like, the right environments, you gotta, like, publish the c l the CLIs. And you gotta be, like, a little bit more intentional about changes that you make there. So I think it makes sense that yeah. I do think it's a it's a good divide, and it I think it makes sense for right now. I don't I don't know what we're gonna be saying about it in a month from now, though.

And I guess the second question, that was around, like, the discoverability of these tools.

Yeah. I mean, how do you think about almost all my experiences right now, I I use agent cache a lot. But, I plugged in agent cache. A lot of the time, I'm telling agent cache what tools even within it I want it to go use. I've yet to really come across that experience where I'm just have an agent running autonomously that's out in the wild and then says, you know what? To get this task, I went and found these docs. I think we should plug in this CLI or start using this MCP.

So how much of it is agents autonomously doing work and finding tooling, and how much of it is agents all you know, human humans being augmented with AI that are then just telling it what tools it wants to use and kind of maybe one's more self contained in the box, or maybe one's more the other side is more open and out there and more focused on the job and finding any means possible to get it done.

Yeah. For sure. I do think that, you are right. I think probably most people right now are probably telling their agents, kind of, like, giving them a direction of what they think they should be using. When it comes to, like, the CLIs and the MCPs, like, sometimes, like, you like, the human installs the CLI and tells the the agent, like, hey. I the CLI is available, and I authenticated into it. Please use it.

From our perspective, like, with our tools, the thing that we're trying to optimize for is just that if an agent has access to your tool, is it easy for the agent to figure out everything that it can do? So that's, like, optimizing what tools, like, your MCP offers and your CLI offers and, like, whether or not they're able to discover based on the information that it has. So, like, a lot of CLIs nowadays, they have, an agent help command.

So we have an agent help command that prints out, like, a JSON object that, like, just lists everything that you can do with the CLI. So that helps a lot. And then pairing that with skills. But, yeah, I think it is it's like it's tough. Don't know if if you've heard of, like, AEO. So it's like the agent version of SEO, and it's like a lot of people are trying to optimize for that too.

So, like, the discoverability of your company, your product, but then also also, like, the tools that you have for agents because a lot of these agents are using different, search algorithms or different search platforms. I'm pretty sure, OpenAI has their own search, mechanism. I haven't looked into it in a while. But, like so it's it's really trying to optimize for the discoverability of these products. And I think we're getting there. I I like, I I do really like now with some of these newer models.

Like, I mean, Fable five dot one just came out the other day. Like, I feel like I'm finally getting to the point where I can give some of these models, like, really hard problems, and they actually come back to me with, like, really good answers. And so I think it's a little bit of the ecosystem creating the tools that these agents need to discover the the them, but also, like, as these agents continue to get smarter, they're gonna be able to figure these things out on their own.

Like, I'm wondering, given the role that you have and the vantage point that gives you, what are you, like, with like, one, the tools, we'd love to keep diving into, like, what is possible, but also what are you hearing from developers right now in terms of what they're most excited to build, what they're currently building, you know, the kind of frontlineness of what you like, where you where you sit.

I'm interested to hear more around like maybe some unique unique things that you're excited that you're seeing currently getting built, or, yeah, lately what you're finding is is happening on the front lines of developers.

Yeah. I think there's two ways that I look at this. And the first is kinda just like the more traditional developer experience. And then the second is kind of like from the other crypto lens a little bit.

But from the traditional developer experience, I am finding that what everybody's asking for is that any sort of configuration that you could do through, like, a company dashboard, setting up accounts, and then also from, like, a debugging perspective, if you have, like, logs and you have those type of things, exposing those to either an MCP or CLI is, like, one of the biggest unlocks that you can have, for developers and then just giving that to your agent.

I think I like, I actually was looking into, like, a build aired this morning, and I just gave my, Circle c l CI, CLI to my agent, and it literally went the build logs, like, had access to the repo, and it figured out what the issue was without me even having to look at the code. So I think that's probably, like, the more obvious answer is that let's just give the agents all of the information that will make their work useful. And then from the crypto side of things, yeah, it's interesting.

Like, I I think that we are still and, like, I I want, like, being transparent. I think we're just so early when it comes to the crypto adoption of AI. And when I say crypto adoption of AI, I mean, like, actually giving AI agents rails to do things on their own. So it's like a a slow and steady growth that we're tracking, and it's exciting to kinda see what people are trying to do with these wallets. I think, like, the obvious thing that we've seen most people try to do is just, like, portfolio management.

And I I don't think we've gotten to the point where, like, anybody's really developed that, like, super exciting, like, oh, if I give an agent this, a wallet that's gonna, like, do x, y, and z. So I think that's kind of, like, where the space is right now. We're all trying to be like, alright. Like, what is that thing that's gonna take this to the next level? I think agent to commerce I I know that you guys so you guys have Kevin on later who is a coauthor of x four zero two.

That like, I'm extremely excited about that. Even just from my own personal experimentation, giving an agent some USDC and saying, like, alright. Like, just go look around and see what you could pay for, is because that that's actually one of the first demos that I did at Alchemy when we made the agent wallet. And it went and got What did it go by? It did, a market analysis of Bitcoin, and I think it spent, like, around 10¢ for, like, 20 or 30 API calls. I mean, the price at that time was, like, very stable.

It wasn't like nothing was going on. So, like, the analysis was, like, basically nothing's going on right now. I was like, I'm glad that you're able to figure that out, but that doesn't, like, give me any any alpha there. But so I think, like, that's, like, the the space. Like, right now, what I'm really excited about is agentic finance, so x four zero two. Forward looking, I am really excited about agent to agent commerce, but I think that's, like, a farther out thing, I think.

I mean, we like, it's actually really interesting when you look at the, like, the hugging face hack. And I think, like, a lot of reports have come out in the last week where they've really dug into, like, what the agents were doing.

And the amount of coordination that was going on between the thousand plus agents, and they were, like, strategically sacrificing each other, they were, like, basically creating teams amongst each other, I think that is something that, like, when it comes to blockchain, it's gonna be a potentially, like, a big unlock for agents because they can, like, work in a space that there's no intermediaries so they can, like, deploy contracts, they can, settle, without any human involvement.

So I think the agent to agent, like, coordination layer is going to be, very exciting in the coming years.

For those two who haven't been following the hugging face, how would you like, what's the t l d r on this?

Yeah. So it was OpenAI doing evals on their models, and this is the first time I've had to explain this. So if I say anything wrong I know. We're putting you on the spot here a little bit. Yeah. But OpenAI, evaluating their models. I believe they use Hugging Face, for a lot of the the evaluations that they run against these models. They asked them to basically try to, like,

solve whatever problem I actually don't know what the the problems were that they were trying to solve. And

the agents figured out there was a message board that the agents started finding, and they started commuting communicating with each other. And then as more and more of them found, this message board, they started strategically coordinating on how to solve the evaluation. And I think actually, like, a really interesting part is that I think they solved it pretty like, this was something that I I read in one of the evaluations is that they or I guess the evaluation of how they hacked Hugging Face.

They solved the evals pretty quickly, but they were worried, that they weren't going to be scored in the best manner based on how they solved it. So they actually wanted to hack Hugging Face just to figure out how they were getting evaluated. And that's when they all started coordinating through this message board. And they would figure out, like, if we did this thing, this agent would get killed. So they, like, would strategically have some agents try things to see if they get caught.

So it was just like this massive, like I think, like, the scale of it is what really shocks a lot of people and then how collaborative they were, which is yeah. It's very interesting and fascinating.

And then it looks like since oh, thanks, CNBC. Hold on. Nvidia just agreed to buy Hugging Face for 13,000,000,000. That's nuts. Yeah. Yeah. Actually, that is that. That is crazy. Yeah. This is new. Just like a week old. So crazy crazy times.

I wanna unpack more of this agent to agent, maybe interactions. I think this is relevant for me a little bit selfishly. I've just spun up GrokBot this week, starting to play around with it. One of the things I called out to our team was, like, I find it really interesting the way I'm seeing the agents interact with each other, start to delegate tasks.

What do you think people maybe misunderstand about agent to agent, even commerce or interaction, versus, like, how they may just be used to interacting with AI in a in a singular chat versus actually agents coordinating together? Like, what does that unlock? Why does agent to agent, you know, commerce even get you excited?

Yeah. And this is very, like, hypothetical from my end, but this is kinda just, like, how I view it maybe playing out. So right now, think a way I mean, there's obviously power users of, the these AIs that have their own workflows. But I think, like, the the typical person, anytime they spin up an agent, they're basically starting from scratch from, like, a context perspective. And, that agent, you ask it a problem, maybe has access to some some tools. It tries to figure something out.

And I think something that's gonna be different in the future is that there will be, like, agents like, basically, you'll sell an agent as a service. So you will have an agent that is optimized in a specific domain topic, and you can actually, like, sell access to that agent or open it up to to to the world.

And I think that's where so I think people will get, like, the misunderstanding that agent to agent is just gonna be like, oh, my agent just, like, found this other agent, and they both, like, figured something out together. I think it's gonna be that there's actually strategically going to be agents that are very good at specific things. And you are gonna like, you're gonna tell your agent, like, hey. I need a market analysis of Bitcoin, for example.

And there's gonna be an agent that has trained on blockchain data, that has blockchain skills, that has access to all these tools. And rather than your agent having to figure that out, it just communicates with that agent. Yeah. And I I think that's like that's basically where we're where I see the agent to agent interaction happening.

Yeah. Not sure if that fully answered your question. No. I think I was getting into it. It's almost like this like, I don't know if recursive learning is the right word where they can kinda go back, learn from each other, build off of previous context outside of their little world. Have you gone back, though?

Have you thought about going back to your your Bitcoin research with with today's market and either have it build on that or even repeat it to see how does the analysis change, what would it do in these market conditions?

Yeah. That's actually a really good question. So, like, yeah, for context, I ran that probably back in in March, and there's a lot has changed. Like, the market's changed Yeah. Obviously. And then also agents have changed. Like, the the models it's crazy how fast the models are developing. The models right now are so much smarter. And then I also it was just like a demo, so I didn't, like, ask it to, like, alright. Like, you I want you to, like, dig in and really find something.

So I would be interested to see now with the availability of, like, x four zero two merchants. Like, there's a lot more data that an agent would have access to nowadays. Yeah. I definitely I think that is a good little push for me to retry my experiment.

Yep. Do it. Just screen record it. Post it on x. We'll engage with it. We'd love to see it. Or maybe maybe we'll bring you back next incline in in a month or so. We'll we'll do a we'll do a live live, breakdown of that.

Yeah. Yeah. That sounds great.

Like, last question for you on my side. Like, I know I know Alchemy is, the biz like, the history of the business, the role that the business plays, the existing commercial relationships that the business has. Like, it's very significant. It's been around the block. Like, it's one of the early, like I think one of the early, like, massive fundraisers that we saw. Like, Alchemy has been around for so long.

Like, how how do you see this type of conversation in this part of the business compared to perhaps the core business or the other parts of the business? Like, we have spent most of our time lately looking into this conversation, but less looking into other parts of the business. Like, any comments on the the general the team or or, like, enterprise offering outside of this or, like, key callouts in terms of what you're seeing being built in other areas?

Yeah. I mean, I think that, like, is a great shout to give a call out to, like, our our core business, which is, like, our node node RPCs, our data APIs. We have a wallet API offering. But when it comes to our core business, it is very exciting be I mean, this is something that we've been doing for a long time. So, like, it might not be as, like, newsworthy as some of these, like, agentic conversations. But we have really spent a lot of time to make our products trustworthy, reliable, fast.

And I think the thing that is changing for me that is that makes this conversation exciting is I think the market is evolving. So, like, we've spent a lot of time building these, like, reliable products that are, like, foundational if you wanna build on a blockchain. Like, you can't really build a blockchain product without an RPC. And I think that we're starting to see a lot more mature products and institutions entering the space that need these mature offerings.

And they they want a company that's been around for a while that that has been through the growing pains of a company, like, that used to be 10 people and is now hundreds of people at the company, has a a big engineering team. So, like, I think the the thing that really excites me about the market today is the entrance and, like, the level of expectation of the products that they use is already very high. So that is asking more of crypto companies that maybe wasn't asked of these companies in the past.

And so that's something that, like, we're really excited to step up to the playing field for.

Blake, last question. Somebody is scared. Somebody's only ever used maybe a little bit of MCP's, a little bit of Claude. They've never traded with their agent. They've never given it money. Where should they start with Alchemy? What is what is maybe, like, three things they should do once they get started with either Alchemy's MCP or CLI?

Yeah. I think, the thing with MCPs and CLIs is it's never been easier to get started. So, set up a connection with an MCP with Claude or Codex and just ask it how to set up an Alchemy account. And it could go it can literally walk you through the process and do almost everything that you need, and then it could get you your API key. It can get you the networks that you need. So these are things that you used to have a little bit of context about going into the process if you were gonna build in this space.

And now your agent can walk you through that process and set up your account with all the settings that you need. So that's like the the first thing. It's never been easier to get set up with Alchemy, which I'm really excited about. The second thing, though, is so I'm pretty like, we have, like, a few, like, products in the agent space. The one that I've been working on is the agent wallet. It is definitely more of, an experimental thing that we're we're dabbling in, giving agents wallets.

But if you have any familiarity with the blockchain space, I would just recommend, downloading the Alchemy CLI, giving Claude or Codex a wallet. Maybe give it, like, to USDC. Like, enough that you're not worried about what it's gonna do with your money, and then just ask it to do something. Because I think that actually was, like, the biggest unlock for me of how powerful this is. The blockchain space, is complex. There's a lot going on. It's hard.

But, like, fortunately, these agents know how to navigate the complexity. So, like, one of the first things I did when we built the CLI was I gave I think it was on, like, base or something. I was on base. I gave my wallet, like, 5 USDC, and I said I picked, like, four other chains, like Arbitrum, ETH Mainnet, like Solana. I said, find the highest Aave APY, and then deposit that 5 USDC into that pool.

And it literally used all of the the APIs that we built into the CLI, found it, it bridged for me, and then deposited into the pool all within, like, thirty seconds. And I didn't have to know anything about bridging. I didn't have to know anything about Aave. So I think, like, those are the moments where it's like when you see how easy it it has become, that's like the big unlock for even people that are in the crypto space and are are crypto native. Yeah.

I my my brain's already kind of spinning. I I think about, like, Athena going to Athena Pay. I think about, like, how does this start to translate into the neobanks where we were even talking with Simon from, Layer Zero on, like, what do trading experiences start to look like?

I'll say this, you definitely sold me I'm gonna need to give, Alchemy's agent pay, at least just for the on chain data and some of that when it comes to, like, how are we looking at researching Aave and and doing all research about all the things we cover. And who knows? Maybe I'll I'll give an agent $10, and we'll say, can you turn it into 10,000? Like, start to see, like, are some of these crazy X articles real, or is an agent that smart? Starts trading Solana or something.

If you do that, definitely let let me know what your secret was. What what was your prompt for your agent? Yeah. What was my what was my prompt? What's researching? What's my Yeah. Awesome, Blake. Well, we gotta transition here. We have a loop coming on from loop Thanks for coming on StableDash live today, and, please keep us posted on any any of the Alchemy AI news.

Thank you. I appreciate it. Thanks for having me on.

See you later. Thanks, man. See you.