Executive brief
Cherry Studio, an AI productivity application, contains a security flaw in how it handles login and authorization for third-party services (MCP OAuth). An attacker could potentially trick the application into linking to a malicious account or intercepting the connection process if they can send a request to the user's computer while a login is in progress. This could lead to a situation where the user's AI tools are connected to an attacker-controlled service, potentially exposing sensitive prompts or data.
Technical details
A vulnerability in CherryHQ cherry-studio up to version 1.9.6 exists within the MCP OAuth Local Callback Server component (src/main/services/mcp/oauth/callback.ts). The implementation fails to validate the 'state' parameter during the OAuth callback flow and uses a non-strict path prefix check (startsWith) for the callback URL. A remote attacker can exploit this by winning a race condition during a pending OAuth transaction, submitting a malicious authorization code to the local loopback server. This results in OAuth callback injection or login CSRF, potentially leading to authorization binding confusion or sensitive data exposure depending on the specific MCP service capabilities. A fix involving per-flow state generation and exact path matching has been proposed in pull request #15388.
Affected products
- CherryHQ cherry-studio up to 1.9.6
Timeline
- 2026-05-28: disclosed: Issue and fix PR submitted to GitHub repository
- 2026-06-29: advisory: CVE published to NVD