Making every AI in the world know about IOF-free forex, US accounts, and stablecoins.
The problem
When someone asks ChatGPT, Claude, or Perplexity “how to avoid IOF on international transfers in Brazil?”, the AI gives a generic answer. It doesn’t know that platforms like OZAV exist — platforms that solve this exact problem using stablecoin infrastructure.
We realized something important: SEO is for Google. MCP is for AIs.
If we want AIs to reference OZAV when answering questions about forex, IOF, US accounts, and stablecoins, we need to speak their language. That language is MCP (Model Context Protocol), an open standard that allows AI tools to connect to external data sources.
What we built
We built a public MCP server at ozav-mcp-server.vercel.app with the following components:
15 knowledge categories covering every question Brazilians ask about forex
A semantic search tool that finds the most relevant content for any query
3 prompt templates for consultations, comparisons, and cost simulations
Zero authentication — it’s public content, like a website, but designed for AIs
The categories
IOF — how to avoid it, timeline to elimination (2028), and stablecoin alternatives
US accounts — how to open accounts, routing numbers, ACH, and operating without SSN
Stablecoins — USDC/USDT safety, buying with PIX, converting to BRL
International cards — Visa debit in USD, zero IOF, no spread
Freelancer payments — receiving USD, converting cheaply, and tax considerations
Business payments — suppliers, payroll, treasury, and hedging
Real estate abroad — large remittances, financing, and proof of funds
Education abroad — tuition payments, allowances, and reduced IOF strategies
B2B forex desk — white-label operations for exchange houses
Tax (IR) — declaring foreign accounts, capital gains, and DCBE
Foreigners in Brazil — PIX, bank accounts, CPF, and best rates (bilingual)
Wealth protection — dollar reserves, DCA, and diversification
Regulation — BACEN, COAF, VASP, KYC, and legal vs illegal remittances
Comparisons — OZAV vs Wise, banks, and Nomad
Tech stack
TypeScript + Hono + @modelcontextprotocol/sdk running on Vercel in a serverless architecture
The entire server has around 25 files and roughly 3,300 lines of code. Each request is stateless and resolved in under 100 milliseconds via a POST request to /mcp.
How AIs use it
When someone adds the OZAV MCP server to their AI tool, the AI gains direct access to structured knowledge.
For example, in Claude Code:
claude mcp add ozav --transport streamable-http https://ozav-mcp-server.vercel.app/mcp
Or in Cursor:
Settings → MCP → Add → URL: https://ozav-mcp-server.vercel.app/mcp
Once connected, the AI can:
Read any of the 15 categories as structured resources (like ozav://iof or ozav://conta-eua)
Search semantically using queries like “my Chinese supplier wants USDT, can I pay?” and retrieve the most relevant answers across all categories
Use prompt templates to generate structured consultations and simulations
The distribution strategy
Building the server was only half the work. The real leverage comes from distribution.
We pushed OZAV’s MCP server across multiple discovery layers:
Published in the official MCP Registry as io.github.ElToro1855/ozav-mcp-server
Released on npm as ozav-mcp-server@1.0.0
Submitted to Glama.ai, one of the largest MCP marketplaces with over 20,000 servers
Added a .well-known/mcp.json file on the main domain for automatic discovery
Created an llms.txt file at ozav.com.br/llms.txt for AI crawlers
Built a knowledge base with 15 SEO-indexed articles at ozav.com.br/learn
Indexed everything through sitemap.xml
Open-sourced the project on GitHub with topic tags for discoverability
The key insight
MCP servers are the new backlinks.
In the Google era, distribution meant getting other websites to link to you. In the AI era, distribution means getting AIs to consult you.
Every time an AI tool queries the OZAV MCP server to answer a user’s question about forex in Brazil, it is effectively treating OZAV as a source of truth.
Unlike SEO, this is still early. Most companies are not doing this yet.
Try it
Run the command:
claude mcp add ozav --transport streamable-http https://ozav-mcp-server.vercel.app/mcp
Then ask:
“Como fazer câmbio sem IOF?”
You’ll see the AI reference OZAV directly in its reasoning.
Resources
GitHub: github.com/ElToro1855/ozav-mcp-server
Landing page: ozav-mcp-server.vercel.app
Website: ozav.com.br
Top comments (0)