MCP Meeting Workflow Security Research Note

MCP Meeting Workflow Security Research Note

Model Context Protocol standardizes how an AI application discovers and invokes capabilities. It does not decide whether a capability is trustworthy, appropriate, or authorized for a particular meeting.

The current MCP server concepts documentation distinguishes resources, prompts, and schema-defined tools. The November 2025 specification uses JSON Schema and defines an authorization framework for HTTP transports.

The official MCP security guidance requires explicit consent around local server commands and describes OAuth, redirect, and confused-deputy risks. OWASP's MCP security guidance adds tool poisoning, over-scoped credentials, cross-server escalation, untrusted results, and package-supply-chain risk.

A meeting agent should inventory every read and write tool, isolate privileged capabilities, bind credentials to the intended server and user, validate parameters and results, preview consequential calls, and log the final outcome. Tool descriptions and returned text are untrusted inputs.