Identity management is getting harder. More applications, more environments and a rapid rise in machine identities are stretching already thin admin teams. Manual processes do not scale, and the longer routine identity tasks take, the greater the operational and security risk.
AI is starting to change that. When paired with the right framework, it can automate repetitive work, reduce manual effort and make access decisions more consistent. One of the key building blocks enabling this is Model Context Protocol (MCP).
For identity security teams, MCP is quickly becoming the standard way AI agents connect to real systems. It allows identity teams to expose capabilities such as user provisioning, policy management and reporting as structured tools that AI can safely use. But this also raises the stakes. The same APIs and credentials used by admins are now being accessed by AI agents, which makes secure implementation essential from day one.
AI operational benefits
Identities, human or NHI, can have their access managed at scale without relying on manual support. That can include developing behavioral baselines, allowing user access based on contextual signals such as device, location or office hours. Any anomalies can be flagged automatically, either blocking access or requesting additional factors before authenticating.
Identity lifecycles can be managed with AI, reducing the risk of breaches by minimizing risks from standing privileges. Users can gain access based on their current role and attributes, rather than simply being granted ‘set it and forget it’ access when they join the organization or move roles.
There’s also a boost in productivity. The adoption of Copilot’s generative AI was found to be ‘associated with a 30.13 percent reduction in security incident mean time to resolution’. AI’s ability to learn from historical decisions and outcomes ensures there can be continuous improvement for business-wide processes. For example, improving accuracy for policy-based access, helping teams keep policies updated to meet compliance requirements. In practice, this does also rely on businesses having the necessary data quality and granularity.
Greater granularity for configurations
Adding context-aware identity administration allows for greater security. For example, identities in select roles, that have specific attributes or have access to sensitive applications can level up authentication requirements. That way, low risk users with expected behaviors can login and access what they need without excessive authentication steps, while friction can then be reserved for higher-risk login attempts.
For example, this could involve asking users for extra biometric authentication, isolating or quarantining entities, or terminating sessions. AI can deliver the automated responses that allow these controls to be implemented consistently, minimizing lateral movement from threat actors and limiting malicious activity.
Supporting admins to do more
Admins can move away from manual identity administration. With more dynamic and contextual security, AI can help make decisions on a holistic basis, something that would be impossible for a human to do at enterprise volume and hundreds or thousands of users. Plus, the rise of NHIs adds even more complexity to identity-related workloads. Especially because they’re likely to operate autonomously, require higher levels of privileges and outnumber human users by 82 to 1.
The solution for admins is to harness AI across multiple use cases. Access provisioning and revoking can be automated, and based on parameters such as activity, attributes, context or role. This supports identity lifecycle management, with automatically updating permissions when an identity or user exits the organization or changes roles. Risks from identities’ behaviors can be assessed with real-time data across networks, flagging anomalies and remediating where necessary.
Further gains come from this level of granular expertise being available to non-technical users, junior admins, or senior admins new to the business. Less time is needed for onboarding, with AI able to generate reports and dashboards on demand. With the right platform, these can be made available through a single pane of glass, for contextual access that informs strategy. This centralized visibility is available with OneLogin, where AI and LLM combine to allow greater automation, integration and protection.
How to leverage AI and LLMs with OneLogin
The OneLogin Model Context Protocol (MCP) server allows admins to query and configure security and authentication policies, automate and accelerate identity administration and access, with commands given through natural language.
MCP was launched by Anthropic in late-2024, as an open standard protocol allowing adaptation and integration between business tools and AI and LLMs.
The OneLogin Model Context Protocol: How it works
Connecting AI clients, such as Claude Desktop, directly to the OneLogin API unlocks 100+ potential actions. AI clients can then manage identities, authenticate and secure the business, and run configuration commands. Plug an AI account into an MCP server, connect to your OneLogin tenant and then admins can run configuration commands.
Users can complete actions and make requests with natural language, opening and accelerating identity administration to multiple teams. These bulk actions can adapt and update based on changing conditions, minimizing the risks of standing privileges and orphaned accounts that may compromise or widen the attack surface.
Key features include monitoring of operations, controlling authentication, and managing identities (users, roles, groups, mappings) plus applications (apps and connectors). These tools act as a bridge for managing all parts of OneLogin, without requiring high-level expertise or lengthy training. Three essential components are at the MCP’s core:
- Host
The AI that wants to engage with external data resources and systems - Client
The interpreter of requests from host to external resources and systems - Server
The external resources and systems being accessed
Examples for OneLogin admins
Anything admins can do via API, the MCP can adjust. In real-time, consistently and at scale. By using natural language prompts, a designated admin can streamline operations with AI to:
- Provision new users and identities
- Get an overview of locked users in production
- Analyze accounts created within a specified time period
- Assign access to multiple users based on complex if/then conditions
- Gather reports on access based on roles, attributes and policies
- Email users at any address stored on their profile
- Create new roles and auto-assign relevant applications, rather than a rigid one-size-fits-all approach
- Produce a summary of roles that currently provide access to environments
- Understand which applications use SAML authentication
- Set up rules that limit user entitlements relating to app usage
- Generate MFA tokens for users who are MFA-registered
Anything admins would do via API can be adjusted, simply by asking in natural language. There’s no need to switch between dashboards to analyze an incident, or review logs and hope that no system silos are limiting visibility. There are also multiple security steps to take so that the resulting insights can be acted on.
Ensuring MCP security
The MCP server runs locally, with the business’s own credentials and operating inside the environment. There are admin controls, with tools for risk rules, smart hooks and API authorization. API level permissions are the same as traditional admin actions, so there’s no need for new roles or permission models that may widen the attack surface.
In the short-term, there’s less disruption because the MCP server is layered with existing APIs and processes. Any RBAC or MFA will still be enforced, and there’s no disruption to logs or audit trails. Security leaders can start the road to further automation and achieving more benefits in the longer term too, where identities are becoming more dynamic and fluid.
AI-assisted identity management: Long-term impacts
Further resource savings come from how the MCP can be a write once, use everywhere layer for identity management. The protocol can act as an alternative to using multiple interfaces for each tool and having to separately manage external connections for AI agents. Instead, there are standardized and structured requests and outputs. And for identity-related workflows, this consistency offers automation opportunities around areas such as lifecycle management, user access, directory services and policy admin. Just bear in mind these best practices when using a OneLogin MCP server:
- User lookup tools
Read-only access - Provisioning tools
User management permissions, not super admin - Reporting tools
Should not have write access
Actions and requests stay logged, creating an audit trail that offers reliable data and insights. Naturally, this supports Just-In-Time and Principle of Least Privilege strategies, helping the business with compliance and governance obligations. After all, as Forrester says, ‘CISOs have to act now’ and that it’s time to treat AI agents ‘as a new identity class. Secure data provenance, memory and enclaves.’
Continuous oversight and visibility help to remove the need for diverting team resources, maybe at short notice, when a periodic audit is due. Across the wider business, teams find more time for strategic value-added activities instead of rigid workflows and routine tasks.
Next steps for leveraging AI and LLMs with OneLogin
We recommend starting with the public GitHub repository. This contains an overview of tools for supported OneLogin API endpoints, plus a short four-minute demo video. The insights will be relevant for OneLogin admins, technical leads and practitioners. You can also get more detailed information that’s customized to your business, just by reaching out to the OneLogin team.
