Production MCP Server for a SaaS Data Platform

Designed and shipped a production Model Context Protocol (MCP) server exposing a SaaS platform's live data to AI agents — turning an existing product into something AI assistants can query and act on directly, not just read about.

Designed the MCP tool surface: defined which platform capabilities and data should be exposed to AI agents, and how, balancing usefulness against attack surface

Built the server on GCP with a Postgres-backed data layer, structured request/response schemas, and webhook-driven updates so agents always see current data

Implemented authentication and access control appropriate for a multi-tenant SaaS context

Delivered working integration with Claude Desktop and Claude Code, with documentation for extending to other MCP-compatible clients

Outcome: the platform can now be queried and operated on directly through AI agents — no custom UI required for AI-native use cases. This is the same class of build I do for clients: taking a real product or internal system and giving it a production-grade MCP interface.