imsg-mcp gives an AI assistant access to someone's private messages and the ability to text from their number. We take reports seriously.
Please do not open a public issue for security problems. Instead, use GitHub's private vulnerability reporting: the repository's Security tab → Report a vulnerability.
Include what you found, how to reproduce it, and the impact you expect. We aim to acknowledge reports within 3 business days and to ship a fix or mitigation for confirmed issues within 30 days, and we will credit you in the release notes if you wish.
- Bypassing authentication or the Access JWT check at the origin.
- Sending a message that the send rules (role, allowlist, length, rate limit) should have refused.
- Reading chats outside
read.scope, or message text reaching logs other than the audit log's record of sends. - Escalating from the bot user to the permissions held by the signed binary, or from a
readtoken to sending. - Anything that weakens the guarantees described in docs/security.md.
- A model being persuaded by message text to ask to send something. That is the known prompt-injection risk the design contains with approval prompts, the allowlist and the rate limit; it is described in docs/security.md. A way to make the server send without passing the rules is in scope.
- Vulnerabilities in
imsg, Cloudflare, macOS or Claude clients. Please report those to their maintainers (forimsg: https://github.com/steipete/imsg).
Security fixes go into the latest release.