← All advisories
CVE-2026-130766highSecret exposure2026-11-18

Obot MCP platform: malicious remote MCP servers could steal users' OAuth tokens, plus SSRF and init-container command injection (CVE-2026-130764/130766/130767/130768, fixed 0.27.0)

Four CVEs for Obot, the open-source MCP gateway and agent platform, were published on 18 November 2026; all are fixed in v0.27.0. The worst lets a malicious remote MCP server name another service in its OAuth protected-resource metadata so Obot fetches the victim's token for that service and sends it to the attacker. The others: SSRF through the OAuth token endpoint a remote MCP server advertises (error bodies from internal hosts are returned), shell command injection via an EOF line in file-type MCP server config values on the Docker backend, and SSRF-guard bypasses using 0.0.0.0, :: or IPv6 transition addresses.

01.

What it is

CVE-2026-130766 / GHSA-x3hf-v55x-jcp6 (Obot v0.26.0 to v0.26.3): Obot only checked that the RFC 9728 resource field existed and did not compare it with the server. The attacker also needs an authorization server that accepts Obot as a client (open registration or client ID metadata documents) and honors RFC 8707 resource indicators, and the victim's consent. Obot doesn't send resource indicators to Microsoft Entra ID.

CVE-2026-130764 / GHSA-57qx-69p6-3w3g (v0.8.0 to v0.26.3): token exchange and refresh used an unrestricted HTTP client that followed redirects, so a Power User who registers a remote MCP server could aim them at cloud metadata or private hosts and read non-2xx response bodies in the OAuth callback error.

CVE-2026-130767 / GHSA-2mfp-9vv3-q737 (v0.9.0 to v0.26.3, Docker backend only): file-type config values were written into a heredoc ending in EOF. A value containing a line with just EOF runs the rest as shell commands in the unprivileged init container, which can change the MCP server's files volume and reach services on Obot's Docker network. The Kubernetes backend is not affected.

CVE-2026-130768 / GHSA-r3m3-5rmh-h3fm (v0.14.0 to v0.26.3): the outbound checks didn't block 0.0.0.0, :: or NAT64/6to4/Teredo/IPv4-compatible addresses, and registry icon URLs were fetched with no egress checks. These are blind requests; Obot reports only status code and content type.

02.

How to check

  1. Check your Obot server version (image tag or the version shown in the admin UI); everything from v0.8.0 up to v0.26.3 is affected by at least one of these.
  2. Review which users have the Power User role or higher and which remote MCP servers that use OAuth have been registered.
03.

How to fix

  1. Upgrade Obot to v0.27.0 or later.
  2. Revoke and re-issue OAuth tokens that users obtained through remote MCP servers you don't fully trust.
  3. Limit the Power User role to trusted people and block Obot's egress to 169.254.169.254 and internal networks.
docker ps --format '{{.Image}}' | grep -i obot
04.

Sources

Related on VibeVulns
More advisories for this tool
The weekly email

One email a week. Only what changed.

Every new security hole in AI coding tools, MCP servers, app builders and the stacks they generate (Next.js, Supabase, Node, PHP), each with how to check your app and the fix. No roundups of things you already read.

Free, one email a week. Read the latest issue first if you like. One-click unsubscribe in every email. We never sell or share addresses.