Safeguard
Vulnerability Analysis•

CVE-2026-54030: MCP OAuth Token Theft in LibreChat

CVE-2026-54030 lets a malicious MCP server steal OAuth tokens in LibreChat before 0.8.5. Upgrade to 0.8.5 and review which MCP servers you have connected.

Safeguard Security Team
3 min read

CVE-2026-54030 is a high severity flaw (CVSS 8.0) in LibreChat before 0.8.5. Its MCP OAuth code does not check that the resource value in server metadata matches the configured MCP server URL, so a malicious MCP server can steal access tokens meant for another server. Upgrade to 0.8.5.

What's affected

  • LibreChat before 0.8.5, per the record and the GitHub advisory.
  • Fixed in 0.8.5.
  • Deployments that connect to MCP servers using OAuth. Instances with no OAuth MCP servers have less to lose, though the advisory does not rule out other uses.

How to check if you're exposed

Find your LibreChat version in the running container image tag or in the project package file, and compare it with 0.8.5. Then list every MCP server your instance connects to and note which use OAuth. Pay attention to servers run by third parties or added by users. The weakness needs an attacker controlled server in that list, so an inventory of connected servers tells you how much of the attack surface you have.

What to do now

  1. Upgrade LibreChat to 0.8.5 or later.
  2. Review the MCP servers configured on the instance and remove any you do not recognize.
  3. If an untrusted server was ever connected with OAuth, treat tokens issued during that time as exposed and revoke them at the provider.
  4. Add an approval step before new MCP servers are connected to shared instances.

Why it matters

The class is CWE-346, improper verification of the origin of data. The advisory points to RFC 9728, which requires that the resource identifier in the returned metadata exactly match the URL the client used for the metadata request. LibreChat copied the metadata value into the authorization request without that comparison.

The impact is on tokens, not on LibreChat itself. A token meant for a legitimate server could be directed to a malicious one. The record does not mark the CVE as exploited in the wild. The CVSS vector shows high attack complexity and required user interaction, which is a reminder that the attacker still needs a user to connect to their server.

For your own projects, the free in-browser dependency scan needs no install, and the Gold directory tracks this CVE.

Try it yourself

Click through it at your own pace, or open it full screen.

Quick check

Do I need a new OAuth client after upgrading?

The advisory does not say. If a suspicious MCP server was connected earlier, revoke and reissue tokens at the provider as a precaution.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.