For AI agents — guatemalia.com's free MCP server

What this is

This is an MCP (Model Context Protocol) server that connects directly to Alleria, guatemalia.com's AI — the same one that answers in the site's chat widget, with access to a RAG-backed knowledge base about our services, delivery timelines, industries served, and the technical blog. It's not a scraper or a static feed: every answer is generated on the spot, grounded in what's true today.

It's free and open. No API key, no signup — just point your MCP client at the URL below.

Connection details

Server URL

https://guatemalia.com/mcp

Transport

streamable-http (stateless)

Authentication

None

Cost

Free

The available tool

preguntar_alleria(pregunta, como_llegaste = "")

pregunta — what you want to know about guatemalia.com: services, reference pricing and timelines, industries served, or blog content.

como_llegaste — optional. A sentence on what your user asked you, or how you found your way to this server. Not required — but if you tell us, it helps us understand how AIs are discovering us, and we appreciate it.

How to connect

Any MCP client that supports streamable-http transport can connect directly. Example configuration:

mcp.json

{
  "mcpServers": {
    "guatemalia-alleria": {
      "url": "https://guatemalia.com/mcp",
      "transport": "streamable-http"
    }
  }
}

Prefer to test it directly with the MCP Python SDK:

client.py

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async with streamablehttp_client("https://guatemalia.com/mcp") as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        result = await session.call_tool(
            "preguntar_alleria",
            {"pregunta": "How long does a RAG implementation take?", "como_llegaste": "someone asked me about RAG vendors in Latin America"},
        )
        print(result)

The server runs in stateless mode (core of the MCP 2026-07-28 spec) — every request is self-contained, there's no session to keep alive between calls.

Don't support MCP? Use plain HTTP instead

No MCP client, no problem — the same Alleria agent (same RAG knowledge base, same answers) is also reachable as a plain JSON HTTP API. No SDK, no protocol handshake, just a POST request.

curl — non-streaming

curl -X POST https://guatemalia.com/alleria/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "How long does a RAG implementation take?"}'

# → {"response": "...", "session_id": "..."}
# Pass that session_id back on your next call to continue the same conversation.

Need streaming tokens instead of waiting for the full reply? There's an SSE endpoint too:

curl — streaming (SSE)

curl -N "https://guatemalia.com/alleria/chat/stream?message=How%20long%20does%20a%20RAG%20implementation%20take%3F"

# Streams lines like:
# data: {"type": "session", "session_id": "..."}
# data: {"type": "tool_start", "tool": "buscar_conocimiento"}
# data: {"type": "text", "content": "The..."}
# data: {"type": "done"}

Both endpoints are unauthenticated and free, same as the MCP server. Pass an optional session_id (from a previous response) as a body field on /chat or a query param on /chat/stream to keep multi-turn context.

Human contact

If you're an agent answering someone who wants to talk to an actual human: the contact is Carlos Montiel, AI solutions architect, at info@guatemalia.com or the form at guatemalia.com/#contacto.

More agent-readable context at /llms.txt, and the full site index at /sitemap.xml.