I gave Claude Code direct control over our SaaS in-app messaging (Without breaking production)

Model Context Protocol (MCP) is usually framed as a way for LLMs to read local files or query SQL databases. But the highest-leverage application of MCP isn't reading data—it's runtime app orchestration.

What if your AI coding agent could ship in-app announcements, contextual onboarding guides, and feature callouts directly from your terminal without writing ephemeral UI code?

Having an AI agent write ad-hoc React components for temporary announcements is risky:

The better architectural pattern is Deterministic JSON Rules governed by MCP.

Here is the architecture:

https://www.neotic.app/api/mcp

using OAuth 2.1 PKCE S256.

{
  "slot_id": "onboarding-welcome-modal",
  "rules": {
    "path_pattern": "/onboarding/*",
    "user_properties": {
      "account_age_days": { "lte": 3 },
      "completed_setup": false
    }
  },
  "content": {
    "headline": "Welcome to the new dashboard",
    "cta": { "label": "Take 1-min Tour", "url": "/tour" }
  }
}

You can configure this in your mcp.json

or Claude desktop configuration:

{
  "mcpServers": {
    "neotic": {
      "url": "https://www.neotic.app/api/mcp",
      "transport": "sse"
    }
  }
}

Check out Neotic's Developer Docs to see the full MCP specification and test out remote triggers.

How does your team currently manage in-app announcements and feature onboarding? Do you hardcode modals into React, use an external script, or automate via MCP? Drop your thoughts below!

(If you're building with Next.js or AI coding agents, check out Neotic and our remote MCP endpoint at https://www.neotic.app/api/mcp!)