Agent Identity & Security
What is the architecture for MCP tool authorization in Catalyst?
Catalyst inserts a policy enforcement point between the agent runner and MCP tools. Each agent has a workload identity from the platform. When the agent calls a tool, Catalyst intercepts the request, evaluates the policy (which agent can call which tool), and either allows or denies it. The policy is defined declaratively and stored in Catalyst's control plane.
Was this article helpful?
Your feedback helps improve Diagrid's FAQ experience.
Keep reading
More Diagrid FAQ articles
- Agent Identity & Security
What is a workload identity for an AI agent and why can't I just use a shared API key?
Explains workload identity for AI agents and the security limits of shared API keys in production.
- Agent Identity & Security
How does Catalyst assign a workload identity to an agent runner like LangGraph or CrewAI?
Describes how Catalyst injects workload identities into framework-agnostic agent runners.
- Agent Identity & Security
What security problems do shared API keys cause when multiple agents run in production?
Lists security problems from shared API keys in multi-agent production environments.