How Lemonade Transformed Its MCPs Into a Governed AI Operating Model for 1700+ Employees With Runlayer

Lemonade is a technology-first insurance carrier offering renters, home, pet, car, and life insurance across the United States and Europe. Powered by AI and social impact, Lemonade replaces brokers and bureaucracy with bots and machine learning, aiming for zero paperwork and instant everything. A Certified B-Corp, Lemonade gives unused premiums to nonprofits selected by its community during its annual Giveback.

Lemonade was experimenting with LLMs long before generative AI entered the business mainstream. For CISO Jonathan Jaffe, agents marked the next phase of evolution, equipping employees to build internal tools, connect company systems, and automate more day-to-day work.

However, as clients such as Claude Desktop put agentic capabilities in the hands of more employees, MCP-powered workflows multiplied across endpoints and quickly outpaced his team’s ability to inventory or govern them. “Suddenly, we went from having a manageable number of people experimenting with agents to skyrocketing usage,” Jonathan shares. “The security team was simply overwhelmed at that point.”

That expansion exposed a central visibility gap that threatened the momentum behind Lemonade’s AI adoption. 1700+ employees could install MCP servers from unverified sources, leaving Jonathan’s six-person team without a registry of what was present, who owned it, which version was active, or what data each connection could access. Investigating each connection individually would consume scarce security engineering capacity, while broad restrictions would interrupt AI-assisted workflows already becoming part of day-to-day operations.

For a regulated insurer handling personally identifiable information, the blind spot also raised the risk that sensitive credentials or protected data could reach an LLM outside Lemonade’s approved controls. The team would then need to determine what was exposed, what remediation was necessary, and whether the affected workflow could continue safely.

Creating and managing a bespoke internal solution wasn’t a good use of time for Jonathan’s team. “Trying to build a service that can handle the number of sessions we get per second and apply good security controls without breaking things was not something that made sense for us to take on,” he explains.

Lemonade needed an enterprise AI enablement solution that would process high session volumes, make accurate security decisions without adding latency or breaking workflows, integrate with its existing stack, and keep Runlayer matched that requirement set, combining the control layer Jonathan needed with a product team moving at Lemonade’s pace.

Runlayer’s integration with Lemonade’s existing MDM system allowed onboarding its managed endpoint population without individual installation or configuration. Runlayer began observing MCP activity across those devices while employees continued working in their existing AI clients.

“Deploying Runlayer was easy,” Jonathan explains. “We simply pushed it out through our MDM, and within a day everybody was onboarded. It just works.”

Once deployed, Runlayer’s endpoint discovery surfaced MCP configuration artifacts across managed devices and revealed a shadow footprint in the low hundreds. Now, Jonathan has clear visibility into which MCP servers employees have installed and where ownership, review, or policy intervention is required.

This live inventory anchors Jonathan’s security decisions in observed behavior. His team now uses Runlayer’s activity data to understand how MCP servers and tools are used across the company, identify requests that cross Lemonade’s risk threshold, and shape policy around the day-to-day workflows those connections support.

Jonathan now uses Runlayer as the distribution and control point for approved MCP servers. Lemonade publishes company-built and externally sourced MCPs through a central marketplace with ownership, versioning, review, and security controls, while managed requests pass through a shared gateway. Employees access reusable capabilities from their preferred AI clients, and Jonathan’s team governs usage from a single control point rather than treating each endpoint connection as a separate security process.

Jonathan has also introduced targeted controls around potential data exposure. These controls inspect MCP requests for sensitive credentials, such as API keys, that may be included unintentionally. Where enforcement is active, they prevent that information from reaching the model. Jonathan’s team focuses intervention on activities that cross Lemonade’s risk threshold while routine MCP use continues.

Runlayer’s responsiveness and willingness to solve complex problems have kept the partnership moving. Jonathan describes its development pace as “faster than what I’ve seen with almost any company I’ve worked with.” That speed matters as Lemonade’s deployment expands and new agentic patterns emerge.

Partnering with Runlayer enabled Lemonade to turn fast-growing endpoint MCP use from an expanding unknown into a managed environment. Today, employees access approved MCP servers through a shared distribution path, while Jonathan’s team supports that use from one central view, refining policy without reviewing every connection individually.

• 1700+ employees connecting MCPs through Runlayer
• 300K+ MCP tool calls governed in 90 days
• 25M+ security checks performed in 90 days

Looking ahead, Jonathan sees a similar governance need emerging as developers and non-developers create and share more skills, plugins, and other agentic capabilities. He plans to extend the principles Lemonade established for MCP servers, including central distribution, ownership, versioning, and review, as those technologies mature.