Yelp Builds a Trust-First Discovery Layer on OpenMetadata for an Agentic Workplace

AI agents are only as good as the data they can find. As agents spread across Yelp, its Analytics Engineering team hit a discovery problem: a catalog crowded with duplicates, thin on documentation, and blind to the trust signals that separate the right table from any table.

Learn how Yelp turned OpenMetadata into an open context layer — the trusted, governed source its people and its AI agents both draw on. The team scaled from a proof-of-concept to a production deployment of roughly 100,000 assets, then engineered a custom search layer, persona-based policies, and a token-efficient in-house MCP server, Yelp's first upstream OpenMetadata contribution. Inside this case study:

• How Yelp customized OpenMetadata's search to weight usage, tiering, and certification, so the canonical, best-documented table rises above a wall of duplicates for people and agents alike
• The read-only, in-house MCP server Yelp built and contributed upstream — improving token economy on core search by ~66% and cutting large-context payloads by ~80% to keep agents fast and reliable
• Persona-based filtering with policies and tags across a ~100k-asset production deployment, so analysts see the data they need while pipeline plumbing stays out of view