MCP is making it remarkably easy for AI agents to connect with our tools, data, and systems. That opens up incredible possibilities but it also creates questions many organizations are only beginning to ask: • Who approved this MCP server? • What can its tools access? • Can it make changes or trigger actions? • Who is responsible if something goes wrong? • And do we have enough visibility to investigate?
These questions led us to work on the 𝗢𝗪𝗔𝗦𝗣 𝗠𝗖𝗣 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 & 𝗥𝗶𝘀𝗸 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸. The framework is designed to help organizations move from “Can we connect this MCP server?” to “Should we connect it and under what controls?”
It provides practical guidance for discovering MCP servers, classifying their risk, defining ownership, applying least privilege, establishing meaningful human approvals, collecting audit evidence, and managing Shadow MCP deployments. But this should not be a framework created in isolation. It needs the experience of people building, securing, governing, testing, and using MCP in the real world.
I would genuinely appreciate your review and feedback. What is missing? What feels difficult to implement? What would make this more useful inside your organization?
Please explore the project, raise an issue, submit a pull request, or share your thoughts with us: 🔗 𝗢𝗪𝗔𝗦𝗣 𝗠𝗖𝗣 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 & 𝗥𝗶𝘀𝗸 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 - https://github.com/OWASP/OWASP-MCP-Governance-and-Risk-Project
MCP adoption will continue to move quickly. Our responsibility is to help ensure that security, governance, and accountability move with it. 𝗟𝗼𝗼𝗸𝗶𝗻𝗴 𝗳𝗼𝗿𝘄𝗮𝗿𝗱 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗳𝗿𝗼𝗺 𝘆𝗼𝘂𝗿 𝗶𝗻𝗽𝘂𝘁𝘀 𝗮𝗻𝗱 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘁𝗵𝗶𝘀 𝘁𝗼𝗴𝗲𝘁𝗵𝗲𝗿. 🙏
#OWASP #MCP #MCPSecurity #AIGovernance #AgenticAI #AISecurity #AppSec #GRC #OpenSource Anthropic Model Context Protocol OWASP® Foundation