AI agents are quickly becoming part of how modern teams get work done, from code generation to infrastructure checks. But letting these tools anywhere near sensitive data like secrets and credentials creates an entirely new set of risks.
Keeper Security’s latest update is meant to address that tension.
The company has launched a new
Model Context Protocol (MCP) integration for Keeper Secrets Manager, giving customers a way to securely connect AI agents, whether they're running locally or in the cloud, with their secrets infrastructure. The goal isn’t just access, but access under clearly defined, auditable rules.
Jeremy London, Director of Engineering, AI & Threat Analytics at Keeper, describes it this way:
“AI agents can only access the specific data and functions they are explicitly authorized for, with every action passing through a secure intermediary that validates permissions in real time and logs detailed events such as
app_client_access and
app_client_record_update.”
That means any request from an AI tool is evaluated, logged, and controlled, and there’s no open-door policy.
MCP also enforces human-in-the-loop oversight for sensitive actions. “Sensitive operations like creating, updating or deleting secrets always require human confirmation,” London explains. “MCP embodies zero-trust and zero-knowledge principles, ensuring full auditability and strict access controls from the ground up.”
Secure Automation Without Sacrificing Control
The integration isn't just about locking down access, it also enables meaningful automation. With MCP, AI agents can generate secure passwords, retrieve or update credentials, manage file attachments, and even run system health checks. These are tasks many teams already do manually or script out with brittle workarounds. Now they can be automated in a policy-driven, controlled environment.
This kind of flexibility is especially relevant for MSSPs and IT teams managing multiple environments. According to London, “MCP is designed for multi-tenant environments, supporting isolated customer profiles, configurable ‘no local logs’ modes and deployment flexibility including on-premises and air-gapped setups.”
MSSPs can configure deployments per client, enforce customer-specific policies, and stream logs to external SIEMs, giving them tight control and visibility without introducing another silo.
Built for Graduated Autonomy and Future AI Use
Long-term, Keeper is thinking about how to balance autonomy and oversight as AI agents become more capable.
“MCP supports a ‘graduated autonomy’ model, permitting AI agents to act independently on low-risk tasks like listing secrets or performing system health checks, while retaining mandatory human approvals for higher-risk actions,” says London.
Admins can define auto-approve rules based on action type, agent identity, or time of day - allowing for flexible, situational access.
Policy-as-code is also in development, giving teams the option to enforce dynamic, contextual rules. That might mean allowing a dev assistant to auto-generate passwords in staging environments during work hours, while flagging production access for review.
What makes MCP different from other integrations is that security isn’t layered on top,it’s core to how the protocol works.
“Every request is validated, policies are strictly enforced and tamper-evident audit logs are generated in real time,” says London. “The MCP server acts as a zero-knowledge intermediary, never exposing encryption keys while maintaining full context for each action, who accessed what, when and why.”
By giving organizations granular control, real-time auditing, and the ability to plug AI agents into workflows without handing them unchecked access, Keeper’s MCP is taking a pragmatic, security-first approach to automation. For teams ready to explore what AI can do, without putting secrets at risk, it’s a timely move.
