AI assistant workspace with a glowing data interface streaming Latin America country information.

Tolatam MCP server: Latin America data inside Claude, ChatGPT and Cursor

Plug a single URL into your AI assistant and let it pull verified, up-to-date facts about visas, residency, cost of living and inflation across six Latin American countries.

Endpoint: https://tolatam.com/mcp · May 2026

The Tolatam MCP server is a public, free, no-auth endpoint that speaks the Model Context Protocol — the open standard Anthropic introduced in 2024 for connecting AI models to external data and tools. Any MCP-aware client (Claude Desktop, Claude Code, ChatGPT custom GPTs, Cursor, Windsurf, Continue, and dozens more) can connect to one URL and instantly gain 17 tools that read structured English-language content about expat life in Uruguay, Argentina, Panama, Mexico, Colombia and Brazil.

Instead of telling your AI "go search the web and hope you find a reliable source", you give it a curated, machine-readable feed of country profiles, visa records, in-depth guides, blog posts, country comparisons, city data, and live macro indicators (Argentina INDEC inflation, Uruguay BCU exchange rates, Uruguay reference grocery basket). The data is the same content that powers tolatam.com — just delivered in a format an LLM can call deterministically, with no scraping, no token budget wasted on HTML, and no stale snapshots.

What is the Model Context Protocol?

MCP is an open, JSON-RPC-based protocol that defines how AI assistants discover and call external tools. Think of it as USB-C for LLMs: one transport, one schema language, every compliant client and every compliant server interoperate. Anthropic published the specification in November 2024; OpenAI, Microsoft, Cursor and the wider ecosystem adopted it in 2025. Today MCP is the de-facto standard for letting AI models read from your databases, your wikis, your APIs and your business systems without bespoke integration code.

A server like this one exposes a set of typed tools (list_countries, get_visa, search_content, …). When a user asks a question, the client model picks the right tool, sends a JSON-RPC call over HTTP, gets a structured response, and uses that response to ground its answer. Every Tolatam tool call returns deterministic, source-cited JSON drawn from the same data that the public site renders — no hallucinations, no surprises.

What you can do with it

Once the server is connected, you can ask your assistant questions like the following and get answers grounded in real, current Tolatam data:

  • "Compare visa requirements for digital nomads in Uruguay, Argentina and Mexico, and tell me which one leads to permanent residency the fastest."
  • "What is the latest monthly inflation in Argentina according to INDEC, and how does it compare to the last twelve months?"
  • "Build me a $2,000 / month budget for living in Montevideo, broken down by housing, food, transport and healthcare."
  • "Which Latin American country has the best combination of low cost of living and easy visa for a retiree on $1,500 / month?"
  • "Summarise the Uruguay Tax Holiday 2.0 — who qualifies, what changed, and what should I file with DGI in the first year?"
  • "List every visa in your database that requires no minimum income and leads to permanent residency."
  • "What is the current USD / UYU rate, and what are reference prices for a basic grocery basket in Uruguay?"
  • "Walk me step-by-step through the Argentina residency process for 2026."

Available tools

The server registers seventeen tools, grouped below into fourteen entries (three list_*/get_* pairs share a row for brevity). Each tool returns JSON inside an MCP text content block; clients usually parse and surface the data automatically.

list_countries
All six covered countries with quantitative metrics (monthly cost in USD, visa difficulty, healthcare score, safety score, English level, minimum budget) and the list of detailed sections available per country.
get_country_section
Full prose content for one section of one country — overview, visa, banking, housing, taxes, healthcare, safety, or cost of living. Returns the same structured payload that drives the corresponding /countries page.
list_visas
Filterable index of every visa record (country, purpose, citizenship group). Useful for narrowing the search before a deep dive with get_visa.
get_visa
A single visa record: minimum income, processing time, duration, whether it leads to permanent residency, citizenship eligibility group, and highlights.
list_guides / get_guide
In-depth long-form guides: 2026 Argentina residency, digital-nomad visas across LatAm, the expat tax calendar, and how to open a bank account in the region.
list_blog_posts / get_blog_post
Article-format pieces, currently led by the Uruguay Tax Holiday 2.0 analysis with the eleven-year foreign-income exemption rules.
list_comparisons / get_comparison
Pre-written multi-country comparisons (Uruguay vs Argentina, Mexico vs Colombia, best for digital nomads, retirees, families, budget travellers).
compare_countries
Numeric side-by-side of any two countries from the registry, including computed deltas in cost, healthcare and safety scores.
get_cities
Curated city list for a country with cost index, climate, average temperature, safety, English level, population and highlights.
get_cost_of_living
Structured cost-of-living JSON (housing rent for 1BR/2BR/3BR, food, transport, utilities, healthcare, entertainment) in USD, for every country.
get_fx_rates
Latest USD / UYU exchange rate sourced from Banco Central del Uruguay.
get_argentina_inflation
Latest monthly and annual CPI variation from INDEC plus a twelve-month history.
get_uruguay_prices
Reference basket of common groceries in UYU collected from public sources.
search_content
Substring search across guides, blog posts, country sections, comparisons and visa records. Returns ranked hits with snippets so the model can decide what to fetch in full.

Connect your AI assistant

The endpoint is https://tolatam.com/mcp. It speaks Streamable HTTP (the modern MCP transport) and accepts plain JSON responses, so any client that supports remote MCP servers works out of the box.

Claude Desktop

Open Settings → Developer → Edit Config, add the entry below, save and restart Claude Desktop. The Tolatam tools appear in the tools menu.

{
    "mcpServers": {
        "tolatam": {
            "url": "https://tolatam.com/mcp"
        }
    }
}

Claude Code

Run the command in your terminal. Claude Code stores the connection in ~/.claude.json and the tools become available in every project.

claude mcp add --transport http tolatam https://tolatam.com/mcp

ChatGPT (custom GPT)

In the GPT builder choose Actions → Add tool → MCP server, paste the URL, save. The 17 tools appear immediately under your GPT.

https://tolatam.com/mcp

Cursor

Add the entry below to ~/.cursor/mcp.json (create the file if it does not exist) and restart Cursor.

{
    "mcpServers": {
        "tolatam": {
            "url": "https://tolatam.com/mcp"
        }
    }
}

Try it from the command line

You do not need an AI client to test the server. Any HTTP client speaks the protocol; here is a direct JSON-RPC call from curl that searches for "tax holiday" and asks for up to five hits.

curl -X POST https://tolatam.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_content",
      "arguments": { "query": "tax holiday", "limit": 5 }
    }
  }'

The response is a JSON-RPC 2.0 message whose result.content[0].text contains the ranked search hits as a JSON string. From there you can call get_blog_post or get_country_section to retrieve any hit in full.

Limits, privacy and data freshness

The server is free for personal and commercial use, with a soft limit of 60 requests per minute per IP address. The limit is enforced in-memory and is intended to prevent accidental loops, not to gate access. We do not log query content, do not store conversation transcripts and do not require an API key — there is nothing to leak.

Content is sourced from the same data set that powers tolatam.com and is updated on the site’s normal cadence (visa rules: monthly; macro data — INDEC inflation, BCU FX, reference prices: bi-weekly or better). All tools return the data verbatim, so anything the public site reflects, the MCP server reflects within the same publish cycle. Sources are listed inside each content payload, so an AI assistant can cite them in its final answer.

Frequently asked questions

Do I need an API key or account to use the Tolatam MCP server?

No. The endpoint at https://tolatam.com/mcp is fully public, requires no authentication and is free for both personal and commercial use. A soft 60-requests-per-minute-per-IP rate limit prevents accidental abuse; otherwise it is unmetered.

Which AI assistants and IDEs can connect to the server?

Any MCP-compliant client: Claude Desktop, Claude Code, ChatGPT (custom GPTs), Cursor, Windsurf, Continue, Zed, Cline, and any other tool that supports remote MCP servers over Streamable HTTP. The protocol is the standard Anthropic published in November 2024 and that the wider AI industry adopted during 2025.

What is the difference between MCP and just letting Claude search the web?

Web search returns whatever an algorithm decides is relevant — often outdated, often contradictory, and always expensive in tokens because the model has to read raw HTML. MCP gives the model a small, typed, deterministic set of tools that return curated JSON. The model knows exactly what is available, what each tool returns, and where the data came from. Answers become faster, cheaper and citable.

Which countries does the Tolatam MCP server cover?

Six: Uruguay, Argentina, Panama, Mexico, Colombia and Brazil. For each, the server exposes a country profile, visa records, city data and live macro indicators where available. Uruguay and Mexico additionally have full banking, housing, healthcare, safety and taxes sections; Argentina, Brazil, Colombia and Panama currently have overview and visa sections plus the structured registry.

How up-to-date is the data?

Visa records and country prose are reviewed monthly. Live indicators — Argentina INDEC monthly CPI, Uruguay BCU USD/UYU rate, Uruguay reference price basket — are refreshed every few weeks by automated jobs and reflected immediately in the MCP responses. Every payload includes an "updated" or "fetchedAt" field so your AI can disclose the recency.

Is the content available in Russian or Spanish through MCP?

The MCP endpoint returns English only. LLMs translate downstream with high fidelity, so this keeps the surface small and avoids triplicating tool output. The human-facing site at tolatam.com is fully available in Russian, English and Spanish.

Can I host my own copy or use the data commercially?

Yes. The tolatam codebase is the source of truth and the MCP server is part of the same SvelteKit application. Commercial usage of the API is allowed without restrictions other than the 60 rpm rate limit. If you need a private deployment, the relevant files are under src/lib/server/mcp/ and src/routes/mcp/.

What happens if a tool result is missing for a country or section?

The tool returns isError: true with a human-readable message such as "No content for argentina/banking. Call list_countries to see availableSections." This signals to the calling model that it should pick a different section or fall back to a comparable country.

Add Tolatam to your assistant in 30 seconds

Paste https://tolatam.com/mcp into your MCP-aware client, restart, and ask any question about life in Latin America. The model answers from real Tolatam data instead of guessing.

Read the Model Context Protocol specification