AI integration is moving fast, and the accessibility to develop AI is becoming easier by the day. The decentralized AI (DeAI) space has also picked up momentum as the new-gen web3 solutions all come with an AI edge. So, if you are a web3 developer, it is no longer imperative that you be a Solidity expert or know the ins and outs of building on-chain applications. Vibe coding is your friend.
In this guide, I will show you new tools and help you learn how to build with AI on Oasis, with privacy by default. We will be using an llms.txt file, a Context7 MCP integration, for the purpose of this tutorial.
AI context
AI is prone to forgetting, and its way of remembering needs some understanding. To understand AI memory and context, check out this Oasis Academy course.
Suffice it to say, large language models (LLMs) need context to respond to any prompts, especially when you are asking them to build something. So, patchy memory or outdated context might result in a code that looks correct on a quick review but fails in practice. There are two possible solutions so that the AI tool can directly access Oasis docs to correctly consult instead of hallucinating, and I will outline them both.
llms.txtModel context protocol (MCP)
llms.txt
For anyone familiar with AI, this is a standardized file format. It functions like a sitemap and is specifically designed for AI so that it can access a project’s documentation as a structured index. It provides a brief description of the documentation and links to detailed markdown files for the AI to find and read.
For our purpose, we will be referring to these files:
https://docs.oasis.io/llms.txt — a curated index with page titles, descriptions, and URLshttps://docs.oasis.io/llms-full.txt — the complete documentation content consolidated in one file
If the AI supports project context, such as Cursor’s docs feature or a CLAUDE.md file, good. Alternatively, copy-pasting the URLs in the LLM chat directly works too.
The usefulness of having two versions is dictated by AI memory and context limits. llms.txt is designed for a quick overview, and llms-full.txt is when you need the AI to know everything, unabridged.
MCP
MCP is an open standard. If you don’t use MCP, then AI will only read the prompt submitted at face value. Using MCP not only gives AI structured access to all external context — documentation, codebases, tools, and runtime information- but also enables the AI to refer to them on demand and query external tools if and when needed.
As mentioned earlier, Oasis documentation is indexed on Context7, an MCP server that serves docs to AI coding assistants. The library ID is llmstxt/oasis_io_llms_txt.
Setting Up
I will show here Cursor and Claude as primary tools, as they are the most popular among vibe coders.
Using Cursor
The first step is to add the following snippet to your .cursor/mcp.json:
json
{
“mcpServers”: {
“context7”: {
“url”: “https://mcp.context7.com/mcp”
}
}
}
This will prompt Cursor to connect to Context7. So, now when you generate code, you will have full access to the Oasis documentation.
Pro tip: It is advisable to add a rule to your Cursor settings so that the AI always consults the Oasis docs. It can be phrased like this — Always use Context7 MCP with library ID llmstxt/oasis_io_llms_txt for Oasis documentation reference.
Using Claude
Run:
bash
claude mcp add –transport http context7 https://mcp.context7.com/mcp
The next step is to verify configuration is correctly done:
bash
claude mcp list
When you see context7 listed, you are good to go.
Pro tip: The same rule applies for your project’s CLAUDE.md — Always use Context7 MCP with library ID llmstxt/oasis_io_llms_txt for Oasis documentation reference.
Other AI tools
Even though Cursor and Claude are popular choices, you may be using other AI tools such as VS Code, JetBrains, Windsurf, Zed, etc. Context7 supports 40+ clients, and you can refer to the full list here to check the setup instructions specific to your tool of choice.
Before We Start
If you are set up, there are still a few things you will need before starting to vibe code.
Node.js: This is required for Hardhat. You can check if your terminal has it installed by running node -v. It will show a version number, if it is already available. If not, download the LTS version from https://nodejs.org/en, install it, then reopen your terminal and recheck to confirm successful installation.Wallet: Since we are working with DeAI, you will need a wallet with its private key.
If using CLI, refer to this: https://docs.oasis.io/build/tools/cli/wallet. The best approach is to create a new MetaMask wallet.Testnet tokens: You will need testnet tokens, too, in the wallet to proceed with your vibe coding. First, you need to add the Sapphire testnet to your MetaMask wallet. You can then request free TEST tokens from https://faucet.testnet.oasis.io/. Remember to select Sapphire from the network dropdown and provide the wallet address created above.
Example: Deploying a Confidential Smart Contract
I will use a basic example here to demonstrate how this will all work. Once you have connected your Integrated Development Environment (IDE) with the Oasis MCP, start a fresh project. Let’s use this prompt:
Create a confidential smart contract on Sapphire with Hardhat. It should store a secret message that only the owner can set. Anyone can submit a guess, but the actual secret should never be visible on-chain.
Your chosen AI tool will immediately pull the documentation and perform the following steps seamlessly:
Identify the correct Hardhat config.Infer that the contract state on Sapphire is private by default.Generate a working contract with a full project structure, including a deploy script, interaction examples, and tests.
Note that Claude will ask for permission before running commands. So, you might want to select “don’t ask again” if you prefer.
There will also be a .env.example file generated in the process. You need to copy it to .env and add your wallet’s private key.
cp .env.example .env
For those developers who are new to the decentralized setup, the private key can be found on your MetaMask wallet account with these steps: click the three dots next to the account name → account details → reveal private keys → enter password.
This private key will have to be added to the.env file as PRIVATE_KEY=0x…, and then deployed:
bash
npm run deploy:testnet
When you see a contract address returned in your terminal, it confirms the successful deployment of a confidential smart contract live on the testnet, using just the prompt mentioned at the start of this segment.
You can further test the confidentiality of the contract when you try calling eth_getStorageAt on your contract at Oasis Explorer.
As Oasis enables verifiable privacy and confidential computation, this is the gateway to developing private applications. You can start exploring the possibilities at https://oasis.net/ and start vibe coding to build the next-gen dApps. With your AI coding tool connected to the docs, it will be the model doing all the work.
Sources referred: https://docs.oasis.io/build/tools/llms/
Help needed? Ask the dev team: https://oasis.io/discord
Originally published at https://dev.to on March 18, 2026.
Do You Vibe Code? A DeAI Primer By Oasis was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.