Okta targets AI agent token costs with MCP scoping
| Source: AI News (ainews.com)
Tags: Okta, MCP, AI agents, token costs, identity, access control, agentic AI, enterprise security
Okta proposes filtering MCP tool lists by agent identity permissions before they reach the model prompt — cutting visible tools by 90%+ in some configurations and eliminating the 'tool tax' where agents burn tokens processing schemas for tools they'll never call.
Details
As AI agents connect to MCP servers exposing dozens or hundreds of tools, every model call includes schemas, names, descriptions, and parameters for every available tool — a 'tool tax' that Okta estimates compounds significantly at scale when multiplied across users and call frequency.\n\nOkta's proposed solution filters the tool list at the identity layer before it reaches the model context, using permissions already managed in the Okta dashboard. An agent with access to 5 of 50 tools on a Workspace MCP server receives only 5 tool schemas per call. Internal modeling found this can reduce visible tools by more than 90% in some permission configurations, with proportional token cost reductions — though Okta did not publish absolute dollar figures.\n\nThe approach also closes a security gap: agents that see unauthorized tools in their context can attempt to call them, and rejection after the fact does not recover the tokens already consumed processing that schema. Pre-prompt filtering prevents the attempt entirely. The capability requires administrators to define per-agent tool permissions in the Okta dashboard — adding governance overhead in exchange for cost control and access enforcement.