Model Context Protocol (MCP) allows AI applications to call APIs, and calling APIs requires proper authorization. Fortunately, the protocol takes this into consideration and requires implementations to use OAuth — a well-established standard. But does that mean an MCP client is just another OAuth client that will work with your existing infrastructure, or do you need additional components, special considerations, or enhanced features?

In this talk, I will take a closer look at authorization in MCP and talk about:

• how securing access to the MCP server differs from securing your APIs,
• best practices for using OAuth features such as consent and refresh tokens with MCP,
• whether you need an MCP gateway to ensure proper protections,
• what security concerns and attack vectors you might face when dealing with MCP.