CVE-2026-44653 is a vulnerability in danny-avila's LibreChat (up to and including version 0.8.3) where users with VIEW access to an MCP server can access decrypted admin-managed secrets through the API endpoints GET /api/mcp/servers and GET /api/mcp/servers/:serverName. The API responses include plaintext values for sensitive fields such as apiKey.key and oauth.client_secret, enabling unauthorized disclosure of provider credentials. The vulnerability is classified under CWE-201 (Insertion of Sensitive Information Into Sent Data). A patch is included in version 0.8.4 that prevents returning these secrets to non-owners by redacting or replacing them with placeholders.

The vulnerability allows users with limited VIEW permissions to obtain plaintext admin-managed secrets, including API keys and OAuth client secrets, which can lead to unauthorized access to underlying AI provider services. This compromises confidentiality but does not affect integrity or availability. There are no known exploits in the wild as of the publication date.

Upgrade LibreChat to version 0.8.4 or later, which contains a patch that prevents returning decrypted admin-managed secrets to users without ownership privileges. Additional recommended mitigations include redacting sensitive fields such as apiKey.key and oauth.client_secret from API responses or returning only boolean indicators of secret presence. If configuration editing without re-entry of secrets is required, secrets should be preserved server-side and replaced with placeholders in API responses. Patch status is not explicitly confirmed in the advisory, but version 0.8.4 is stated to contain the fix.