Model Context Protocol (MCP) security vendor Runlayer announced today a $11 million seed round led by Khosla Ventures and Felicis, while simultaneously ending its four-month stealth operation. Founder Andrew Berman revealed that Runlayer has signed eight unicorns or public companies, including Gusto, dbt Labs, Instacart, and Opendoor, and has attracted David Soria Parra, the main author of the MCP specification, as a consultant.

Since the open-sourcing of the MCP protocol in November 2024, it has been adopted by major model vendors such as OpenAI, Microsoft, AWS, and Google to provide AI agents with the capability to read and write data and perform business operations without human intervention. However, the protocol itself lacks security mechanisms, and platforms such as GitHub and Asana have reported prompt injection and data access violations, creating demand for gateways, audits, and permission management.

Runlayer launched an integrated security platform that combines gateways, threat detection, observability, enterprise automation development, and fine-grained permissions into a single console. The product uses an "Okta-style" directory, allowing IT departments to pre-approve MCP servers and map them to employee identities, thereby enabling one-to-one mapping between agent permissions and user permissions to prevent unauthorized reading and writing. The platform supports identity providers such as Okta and Entra and provides full-chain audit logs to meet compliance requirements.

Berman previously served as the Director of AI at Zapier, participated in building early MCP servers, and had deep collaboration with OpenAI and Anthropic. He believes that the rapid deployment of the protocol has created "observability blind spots," leading him and two former Zapier colleagues to leave in August and found Runlayer, completing product development and customer validation within four months. The company plans to expand its engineering team this year and launch the GA version. The next step will be to support on-premises private deployment and multi-cloud environments.