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

Description and use

Using MCP and deterministic JSON rules, an AI agent (Claude Code) controls in‑app messaging for a SaaS product without adding temporary UI code, reducing bundle size and styling drift.

Problem

AI agents generating ad‑hoc React components for temporary in‑app announcements risk bundle bloat, CSS inconsistencies, and SSR layout breaks.

Outcome

Dynamic in‑app messaging is now driven by deterministic JSON rules, eliminating bundle bloat and styling drift while maintaining zero server latency.

Workflow

• 1. Prompt Claude Code or Cursor to announce a release.
• 2. Agent connects to https://www.neotic.app/api/mcp using OAuth 2.1 PKCE S256.
• 3. Agent emits a structured JSON payload with slot_id, rules (path_pattern, user_properties), and content (headline, cta).
• 4. Browser evaluates rules locally via a micro‑SDK (<4KB) with zero server latency.
• 5. Matching rules trigger the in‑app message without adding new React components.

Evidence

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.

Source

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

Source