Outlook + OneDrive Assistant: why it is not safe to use
We found sensitive Microsoft account access and several destructive actions without an evident built-in confirmation step.
What we found
Our review did not find hidden instructions, hardcoded secrets, obfuscated code, or undisclosed third-party data sinks. The inspected code and docs point to Microsoft Graph and Power Automate for the connected Microsoft 365 account, which matches the skill's stated purpose.
The main concern is operational risk. The skill exposes actions that can be destructive or hard to reverse, including deleting OneDrive items, sending email, accepting calendar invites, and toggling or running Power Automate flows. In the handlers we inspected, we did not see an evident built-in confirmation gate before those actions execute.
We also found that the skill reads Microsoft OAuth credentials from environment variables and stores OAuth tokens in a local file under the user's home directory. That is common for this type of integration, but it still increases the sensitivity of the environment where the skill runs.
What to do instead
Use this skill only in a tightly scoped Microsoft 365 account, with least-privilege app permissions and a separate test tenant if possible. Add an explicit human confirmation step for delete, send, invite-response, and flow actions before enabling broad use. Protect the host environment and local token storage, rotate credentials when needed, and prefer alternatives that keep approval prompts and token handling under your existing admin controls.
Source: https://github.com/ryaker/outlook-mcp
We report what our security review found at the time we checked, with the goal of keeping people safe. Projects change; if a maintainer has since fixed this, we are glad to recheck it. Email hello@agentpod.com.