Secure MCP gateway
Control MCP tool access with tenant and workspace identity, scoped credentials, policy checks, approvals, execution receipts, and correlated audit evidence.
The gateway starts with the tenant, workspace, user or service principal, and the agent responsible for the request. Authority is not inferred from prompt text.
Evidence: Tenant and workspace context carried into the request boundary
Workspace MCP assignments and action policy determine which connection and tool may run. High-impact actions can require a human approval before execution.
Evidence: Connection, tool, action class, policy decision, approval state
The execution layer resolves the approved connector credential for the scoped action. The agent receives the result it needs, not a reusable bearer token in its prompt.
Evidence: Credential reference and execution receipt instead of prompt-carried secrets
Correlation identifiers connect the session, policy result, tool call, approval, output, and downstream outcome so teams can investigate failures and prove what occurred.
Evidence: Session, request, tool, approval, result, and outcome linkage
The request identifies the workflow context and the MCP capability needed.
Tenant, workspace, tool scope, action class, and approval requirements are evaluated before a connector is invoked.
The approved credential is resolved at the execution boundary and the scoped tool call runs.
The policy decision, execution status, result metadata, and outcome remain correlated for operations and audit.