We all know AI is amazing, sure, but let’s face the truth: LLMs struggle with Algorand code, often using outdated patterns and deprecated libraries. So, the DevRel team spent the last few weeks working to fix that.
Today, we’re excited to release the result: a collection of Agent Skills and setup guides for MCP tools that, together, give LLMs the context they need to write correct smart contract code and speed up your workflow.
Let’s dig into what that means.
The problem
LLMs are trained on public code, but Algorand's tooling has evolved quickly – too fast for even frontier models to keep up.
When you ask an AI to write a smart contract, it hallucinates deprecated APIs, invents features that don't exist, or misses AVM-specific constraints entirely.
This became obvious at the hackathons we sponsored. AI tools are everywhere now, and participants skip the docs entirely, relying on AI to move fast under time pressure. But we spent more time helping them fix AI mistakes than they saved by using the AI tools.
Agent Skills
Agent Skills are Markdown files that teach an AI how to perform a specific task. They're similar to prompts but structured to greatly reduce the LLM's context load. Your coding agent (Claude Code, Cursor, Copilot, etc.) can reference them when relevant.
For example, the build-smart-contracts skill explains how to write contracts in Algorand TypeScript or Python, what the compiler expects, and common pitfalls to avoid. When you ask your agent to build a contract, it pulls from this skill instead of guessing.
We’ve built over a dozen skills, covering everything from smart contract testing to frontend integration.
MCP (Model Context Protocol)
Skills provide static knowledge, but sometimes an agent needs to look something up. That’s what MCPs are used for.
MCP is a protocol that lets AI agents connect to external tools and data sources. We’ve provided setup instructions for two MCP servers:
-
Kappa MCP – the same backend that powers the AI chat on the official documentation. When your agent encounters something the skills don't cover, it can query our docs directly.
-
GitHub MCP – retrieves canonical code examples from our repositories, giving your agent real, working reference implementations.
How it comes together
With skills providing baseline knowledge and MCP servers enabling real-time research, your agent can handle real development tasks with much better accuracy.
There's another factor working in our favor: Algorand smart contracts are written in TypeScript or Python. LLMs are already super proficient in these languages. We're not teaching them a new language from scratch, like Solidity or Move. We're just teaching them the Algorand-specific exceptions on top of languages they already know.
For example, one skill tells the agent to use uint64 instead of number since that's the type native to the AVM, and to prefer value semantics over reference semantics since that's how the AVM works. With just those two hints, I watched the agent leverage its existing TypeScript expertise to produce very complex smart contracts, reasoning about logic with those caveats in mind.
Teaching the exceptions to the rules, instead of the whole rulebook (a custom language), turned out to be surprisingly effective.
What we're working toward
For experienced Algorand developers, this should reduce friction and increase velocity. Less time fighting the AI, more time shipping.
For newcomers, it lowers the barrier to entry. The agent can guide you and flatten the learning curve using tools they already know.
Internally, our DevRel team is using this to build demos faster. If it works, there's a potential flywheel effect: more developers shipping more Algorand code means better training data for future LLMs, which means even better Algorand support out of the box.
Build with us
Get started today with our github resources https://github.com/algorand-devrel/algorand-agent-skills. We've also created a dedicated #agentic-dev channel in Discord. Come ask questions, troubleshoot your setup, and share what you're building.
Disclaimer: The content provided in this blog is for informational purposes only. The information is provided by the Algorand Foundation and while we strive to keep the information up-to-date and correct, we make no representations or warranties of any kind, express or implied, about the completeness, accuracy, reliability, suitability, or availability with respect to the blog or the information, products, services, or related graphics contained in the blog for any purpose. The content of this blog is not intended to be legal, financial, or investment advice nor is it an endorsement, guarantee, or investment recommendation. You should not take any action before conducting your own research or consulting with a qualified professional. Any reliance you place on such information is therefore strictly at your own risk. All companies are independent entities solely responsible for their operations, marketing, and compliance with applicable laws and regulations. In no event will Algorand Foundation nor any affiliates be liable for any loss or damage including without limitation, indirect, or consequential loss or damage, or any loss or damage whatsoever arising from loss of data or profits arising out of, or in connection with, the use of this blog. Through this blog, you may be able to link to other websites which are not under the control of the Algorand Foundation. We have no control over the nature, content, and availability of those sites. The inclusion of any links does not imply a recommendation nor endorse the views expressed therein.