Okta Scopes MCP Tools to Cut AI Token Costs

Model Context Protocol, or MCP, gives AI agents a standard way to access external data and services. But servers can expose large tool catalogs, forcing models to process lengthy definitions with each prompt even when most functions are irrelevant or unauthorized. Okta says identity-based scoping can curb that “tool tax” while aligning agent access with existing user entitlements and security policies.

Okta has proposed filtering MCP tool lists before a prompt reaches the model, removing functions a user is not authorized to invoke. In internal tests, the approach cut visible tools and the related token overhead by more than 90% in a specific permissions scenario. The company has not disclosed a dollar estimate for the savings or a formal launch date, leaving the reported result as an early benchmark rather than a broad production guarantee.