Pinterest engineering has deployed a production-scale Model Context Protocol (MCP) ecosystem enabling AI agents to automate complex engineering tasks across internal tools. The architecture uses domain-specific cloud-hosted MCP servers (covering Presto, Spark, Airflow, etc.) connected via a central registry that enforces governance, permissions, and service discovery. A two-layer authorization model with JWT and mesh identities, plus human-in-the-loop approval for sensitive operations, ensures security and compliance. As of January 2025, the system handles 66,000 invocations per month across 844 active users, saving approximately 7,000 hours per month. The approach replaces ad hoc integrations with a standardized, auditable substrate integrated into developer IDEs and chat platforms.
Sort: