Descope AI is the identity layer for AI agents and MCP servers from Descope, delivered as the Agentic Identity Hub. It adds authentication, consent, SSO, access control, credential management and policy controls to AI systems. Developers can register APIs as OAuth-protected resources, expose MCP servers with OAuth 2.1 and PKCE, issue short-lived scoped credentials so agents never hold the application's token, and keep a directory of every agent with its users, tenants and tool-level scopes. Audit logs can be viewed in the dashboard or streamed to a SIEM.
It targets engineering and security teams at companies shipping AI products, as well as organizations that run internal agents and must govern them. Developers building B2C or B2B AI apps use it to authenticate users with low-code journeys, while IAM teams use the directory and policies to enforce least privilege. Customer admins can manage agent access to an MCP server through Cross-App Access with their own identity providers.
Typical scenarios include securing a public MCP server so that only approved clients and scopes can call its tools, giving an agent a revocable token to reach a third-party service through one of more than fifty prebuilt connection templates, requiring human approval before an agent performs a sensitive action, spotting shadow agents and revoking their access instantly, and prompting users to reauthenticate during risky steps. It works alongside FastMCP, Vercel, Reflex, Skyflow and Golf.dev, and the parent product also covers general customer authentication. Descope presents the offering as standards-based, so teams can connect existing identity providers instead of replacing them.
