Okta has introduced a new approach to reducing AI agent token costs by incorporating Identity-Scoped Model Context Protocol (MCP) tool lists into their platform. This means that instead of having an AI agent make individual model calls for each tool exposed by the MCP server, it can request schemas, names, descriptions and parameters for every tool in advance.
This approach is known as "tool tax," which refers to the tokens consumed as a model considers tools, including those used for model deployment. By scoping out these tools beforehand, Okta claims to be able to reduce AI agent token costs, making it more feasible to power complex machine learning models without exceeding budget limits. This new approach is just one part of a broader effort by Okta to optimize the cost and performance of their identity management platform.
The MCP tool list concept has gained traction in the AI industry as a way to optimize model deployment and reduce token costs. By using this approach, organizations can unlock the full potential of their AI models without breaking the bank. As more companies adopt this strategy, we can expect to see further advancements in AI technology and the adoption of cost-effective solutions like Okta's MCP tool list approach.