MCP for marketing automation: bring your own AI agent to the email channel

Sendnomi is not an AI provider. We’re explicit about this — choosing your AI provider is a business decision now, not an infrastructure decision. Which model you run with, the data residency, the cost, the prompt privacy — those are yours to know, not ours.

That’s why Sendnomi’s marketing and operational layer runs over MCP (Model Context Protocol). You bring your own AI agent (BYOA — Bring Your Own Agent), and it talks to us via the open standard.

MCP in brief

MCP, released in 2024, is an open standard for AI agents to talk to external tools. An AI agent (an IDE plugin, terminal CLI, or your custom client) connects to an MCP server; the server exposes tools, the agent calls them to do real work.

The Sendnomi MCP server exposes 26+ tools:

• campaign.create — create a new campaign
• campaign.send — send a campaign
• segment.list — list saved segments
• segment.create — create a new segment by criteria
• contact.search — search contacts
• template.render — render a template with variables
• webhook.subscribe — subscribe to events
• analytics.summary — pull performance summary
• …and more

Full list: /en/mcp-server/

Architecture

┌─────────────┐                      ┌──────────────────┐
│  Your AI    │ ◄────── MCP ────────► │ Sendnomi MCP     │
│  agent      │      (stdio/SSE)      │ Server           │
│  (IDE       │                       └─────────┬────────┘
│  plugin,    │                                 │
│  CLI,       │                                 ▼
│  custom)    │                       ┌──────────────────┐
└─────────────┘                       │ Sendnomi API     │
                                      │ (mail engine)    │
                                      └──────────────────┘

The AI agent lives on your side (local or your own cloud); only tool calls go to Sendnomi — the model inference is entirely yours.

Concrete example: abandoned cart campaign

Scenario: send personalized reminders to customers who abandoned a 100+ TL cart on your e-commerce site over 24 hours ago.

1. Your AI agent receives a prompt

“Send personalized reminders to customers who abandoned 100+ TL carts in the last 24 hours. First mail with a 5% discount, second mail 48 hours later with 10%.“

2. Agent talks to Sendnomi via MCP

[AI agent] → mcp call: segment.create
  name: "cart-abandoned-24h-100tl"
  filter: { event: "cart_abandoned", since_hours: 24, min_value: 100 }

[Sendnomi MCP] ← response: { segmentId: "seg_8f3a..." }

[AI agent] → mcp call: campaign.create
  name: "Cart reminder — wave 1"
  segment: "seg_8f3a..."
  template_id: "tpl_cart_v3"
  schedule: { type: "immediate" }
  variables: { discount: 5 }

[Sendnomi MCP] ← response: { campaignId: "camp_2b9d..." }

3. Second wave is scheduled

[AI agent] → mcp call: campaign.create
  name: "Cart reminder — wave 2"
  segment: "seg_8f3a..."
  template_id: "tpl_cart_v3"
  schedule: { type: "scheduled", at: "+48h" }
  variables: { discount: 10 }

The agent generates all of this from a natural-language prompt without you writing any of it.

Which agents?

Typical MCP client-side agent categories:

• IDE plugins — agentic extensions inside your code editor (multi-step tool calling, planning)
• Terminal CLIs — command-line AI clients managing developer flows
• Open-source agentic IDEs — full IDEs with integrated AI orchestration
• Your own custom agent — buildable in ~100 lines with Python/TypeScript SDKs

The Sendnomi MCP server doesn’t care which client you use — it’s an open standard.

Data residency clarity

Tool calls flowing to Sendnomi via MCP:

• Do they go to your AI provider? No. Tool call payloads only go to Sendnomi.
• Does Sendnomi see your prompts? No. The MCP client (the agent) does model inference; only the resulting tool calls reach us.
• Is customer content sent to your AI provider? Only as much as you pass into your AI agent’s prompt + context. Sendnomi doesn’t see that traffic.

Getting started

1. Create a Sendnomi MCP API key — Panel → API Key → new key with MCP scope.
2. Add it to your MCP client — claude_desktop_config.json or .cline/config.json:

   {
     "mcpServers": {
       "sendnomi": {
         "command": "npx",
         "args": ["-y", "@sendnomi/mcp-server"],
         "env": {
           "SENDNOMI_API_KEY": "sk_live_..."
         }
       }
     }
   }

3. Give your agent its first command — “Summarize last week’s campaign performance.”

Detailed docs: /en/docs/api/

Conclusion

To manage email marketing with an AI agent, you don’t have to stay tied to a SaaS’s AI provider. Through the MCP open standard, you keep your own model, your own prompts, your own data residency while using all of Sendnomi’s capabilities.

— Yazılım Koçu