On this page

MCP & Agentic Architecture

Tajo is an agentic orchestration layer for Brevo. Instead of manually building campaigns, segments, and automations, you describe what you want in natural language and AI agents execute across Brevo’s full engagement stack.

Tip

The Model Context Protocol (MCP) is an open standard that lets AI models interact with external tools and APIs. Tajo exposes Brevo’s entire API as MCP tools, so any LLM can orchestrate customer engagement autonomously.

Architecture

Tajo’s agentic stack has three layers:

+--------------------------------------------------+
|        Natural Language Interface                 |
|   "Win back customers who haven't ordered in 90  |
|    days with a 3-step email + SMS sequence"       |
+--------------------------------------------------+
          |
          v
+--------------------------------------------------+
|             AI Agents (Orchestration)             |
|   Campaign Orchestrator | Customer Intelligence   |
|   Win-Back Agent | Cart Recovery Agent            |
|   ─── Agents compose Skills based on goals ───   |
+--------------------------------------------------+
          |
          v
+--------------------------------------------------+
|              Skills (MCP Tools)                   |
|   brevo/send-email | brevo/create-contact         |
|   brevo/track-event | brevo/send-whatsapp         |
|   ─── Atomic, deterministic, API-backed ───      |
+--------------------------------------------------+
          |
          v
+--------------------------------------------------+
|              Brevo API (Engagement)               |
|   Email | SMS | WhatsApp | CRM | Automation       |
+--------------------------------------------------+

Layer 1: Skills (MCP Tools)

Skills are atomic automation units that wrap Brevo API endpoints. Each skill has defined triggers, actions, and permissions. When exposed via MCP, skills become tools that any AI model can invoke.

name: send-campaign-email
version: 1.0.0
mcp_tool: brevo/send-email
description: Send a transactional or campaign email via Brevo

triggers:
  - event: agent_request
  - event: cart_abandoned

actions:
  - brevo/send-email

brevoEndpoints:
  - POST /v3/smtp/email

permissions:
  - email:send

View all Skills | Skills Format Specification

Layer 2: Agents (Orchestration)

Agents are LLM-powered orchestrators that compose multiple skills to achieve a marketing goal. Each agent is defined in a markdown file with frontmatter specifying its capabilities, tools, and constraints.

---
name: win-back-agent
description: Re-engage churned customers with personalized multi-channel sequences
temperature: 0.3
max_tokens: 4096
tools:
  - brevo/list-contacts
  - brevo/create-segment
  - brevo/send-email
  - brevo/send-sms
  - brevo/track-event
---

You are a customer win-back specialist. Given a churn threshold
and customer segment, design and execute a re-engagement sequence.

## Your Capabilities
1. Query customer data and identify churned segments
2. Create targeted segments based on purchase history
3. Design multi-step email + SMS sequences
4. Track engagement events and adjust strategy
5. Report on recovery rates and revenue impact

Agent Specification Format | Building Your First Agent

Layer 3: Natural Language Interface

The top layer translates marketer intent into agent actions. A marketer says:

“Create a win-back campaign for customers who haven’t purchased in 90 days. Start with an email offering 15% off, follow up with SMS after 3 days if no open, then a final WhatsApp message with a personalized product recommendation.”

The orchestration layer:

Selects the Win-Back Agent
Agent queries Brevo contacts via brevo/list-contacts skill
Creates a segment via brevo/create-segment skill
Designs the 3-step sequence using brevo/send-email, brevo/send-sms, brevo/send-whatsapp skills
Sets up event tracking via brevo/track-event skill
Monitors and reports results

Why MCP?

The Model Context Protocol gives Tajo three critical advantages:

Advantage	Description
Model-agnostic	Works with Claude, GPT, Gemini, open-source LLMs — any model that supports MCP
Composable	Skills snap together like building blocks. Agents decide which to use at runtime
Observable	Every tool call is logged with inputs, outputs, and decisions. Full audit trail
Secure	Permission-scoped tools. Agents only access what they’re authorized to use

Getting Started

Set up the Brevo MCP Server

Connect your Brevo account and expose API tools via MCP.

Brevo MCP Server Setup
Build your first agent

Create a marketing agent that orchestrates Brevo skills.

Building Agents Guide
Deploy and monitor

Run agents on a schedule or trigger them from events.

Agent Deployment

Pre-Built Agents

Tajo ships with ready-to-use marketing agents:

Agent	Purpose	Skills Used
Campaign Orchestrator	Design and execute multi-channel campaigns from natural language	send-email, send-sms, send-whatsapp, create-segment
Customer Intelligence	Autonomous segmentation, RFM scoring, churn prediction	list-contacts, create-segment, track-event
Win-Back Agent	Re-engage churned customers with personalized sequences	list-contacts, create-segment, send-email, send-sms
Cart Recovery Agent	Recover abandoned carts with timed multi-channel nudges	track-event, send-email, send-sms
Data Sync Agent	Bi-directional sync between any platform and Brevo	create-contact, update-contact, track-event
WhatsApp Commerce Agent	Conversational commerce via WhatsApp Business	send-whatsapp, track-event, list-contacts

Next Steps

Brevo MCP Server Setup — Connect Brevo as MCP tools
Agent Specification Format — Define custom agents
Building Your First Agent — Step-by-step tutorial
Skills Reference — Browse all available MCP tools