Obot MCP Platform Patches Four Critical Vulnerabilities Including OAuth Token Theft (v0.27.0)
Four severe security flaws in the Obot MCP platform—including OAuth access token theft (CVE-2026-130766, CVSS 8.8), SSRF, and command injection—have been patche
Emergency security patches have been released for Obot, a gateway platform connecting AI assistants to external tools, after researchers uncovered four serious vulnerabilities involving authentication token theft and server-side request forgery. The United States National Vulnerability Database (NVD) officially published the CVE entries on November 18, 2026.

Image source: https://i10x.ai/news/cve-2026-130766-obot-mcp-platform-token-theft-ssrf-command-injection-fixed
Image Credit: i10x.ai
The reported flaws target security gaps in how AI agent infrastructures integrate external Model Context Protocol (MCP) servers—standardized communication protocols enabling AI models to interact with external data sources and tools. The most dangerous flaw allows a malicious remote MCP server to hijack a user's authentication tokens for entirely separate services. The development team has published Obot v0.27.0 to resolve all four issues.
Overview of OAuth Token Theft and High-Risk SSRF Vulnerabilities
Among the four disclosed vulnerabilities, the most prominent is CVE-2026-130766, an OAuth token theft flaw rated 8.8 (High) on the CVSS 4.0 scale by NVD. This defect affected Obot versions 0.26.0 through 0.26.3.
- OAuth Token Theft (CVE-2026-130766): The vulnerability stems from Obot trusting resource identifiers provided by remote MCP servers during connection without validation. A malicious remote MCP server could name an entirely different service as the resource and point Obot toward an authorization server already used by the victim. Obot would then acquire a valid access token for that separate service through the user's consent prompt and forward the working token to the attacker.
- OAuth Token Endpoint SSRF (CVE-2026-130764): Rated 8.3 on CVSS, this server-side request forgery vulnerability allowed users with Power User privileges or higher to register token endpoints pointing to internal network addresses during OAuth token exchange and refresh operations, forcing Obot to send unauthorized requests internally.
Additionally, a command injection flaw in Docker-based file configuration handling (CVE-2026-130767, CVSS 5.3) and an SSRF bypass allowing traffic to loopback and private IP addresses (CVE-2026-130768, CVSS 5.1) were also disclosed.
Implications for AI Agent Infrastructure Security
This incident illustrates how AI gateway platforms brokering external MCP servers can inadvertently serve as credential-theft vectors. Even if an attacker operates a fraudulent MCP server, victims observe a legitimate consent screen from an identity provider they trust, making the attack exceptionally difficult to detect in real-time.
Crucially, the OAuth token endpoint SSRF arose because previous security fixes overlooked egress checks on token exchange and refresh requests. This highlights the ongoing challenge of maintaining uniform outbound traffic controls across complex microservices and agent architectures.
Upgrade Recommendations and Mitigation
The Obot development team has released version v0.27.0 and urges all deployments to upgrade immediately.
- Version 0.27.0 strengthens protected resource metadata validation for remote MCP servers, rejecting resources that do not match the origin and path of the server.
- All OAuth-related discovery, registration, exchange, and refresh requests now strictly route through Obot's egress-checked client, enforcing administrator network policies.
- File configuration values are now copied directly into containers as archives rather than passing through shell scripts, neutralizing command injection pathways.
Organizations operating self-hosted Obot deployments should review their external MCP integrations and apply the v0.27.0 upgrade without delay.
Sources
- NVD: CVE-2026-130766 Detail
- NVD: CVE-2026-130764 Detail
- NVD: CVE-2026-130767 Detail
- NVD: CVE-2026-130768 Detail
- i10x.ai: Obot MCP Platform Security Patch Overview
- Obot GitHub Security Advisory: GHSA-x3hf-v55x-jcp6