Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-101084 Overview
CVE-2026-101084 is an authorization bypass vulnerability in the obot platform affecting versions before v0.21.1. The /mcp-connect endpoint fails to enforce Access Control Rules, allowing any authenticated user who knows a server ID to connect to restricted Model Context Protocol (MCP) servers. Attackers can leverage stored OAuth credentials to invoke MCP tool calls against sensitive backend systems. The flaw is tracked under [CWE-639: Authorization Bypass Through User-Controlled Key].
Critical Impact
Authenticated attackers can bypass access controls on restricted MCP servers, reach backend systems through stored OAuth credentials, and manipulate data via MCP tool calls.
Affected Products
- obot platform versions prior to v0.21.1
- Deployments exposing the /mcp-connect endpoint to authenticated users
- obot instances with configured MCP servers using stored OAuth credentials
Discovery Timeline
- 2026-09-27 - CVE-2026-101084 published to NVD
- 2026-09-30 - Last updated in NVD database
Technical Details for CVE-2026-101084
Vulnerability Analysis
The obot platform exposes an /mcp-connect endpoint that brokers connections between authenticated users and configured MCP servers. Access Control Rules in obot are intended to restrict which users can reach specific MCP servers. In versions before v0.21.1, the endpoint does not evaluate these rules before establishing a connection.
Any authenticated user who can supply a valid MCP server identifier reaches the restricted server. Once connected, the user issues MCP tool calls that execute using OAuth credentials stored by the platform on behalf of the authorized owner. This converts a low-privilege authenticated session into access over sensitive downstream systems.
The weakness is classified as [CWE-639], an authorization bypass driven by a user-controlled key. The server ID functions as the sole access gate rather than being validated against the caller's permissions.
Root Cause
The /mcp-connect handler omits the access control check that gates MCP server usage elsewhere in the application. Authorization relies on knowledge of the server identifier rather than on an explicit permission lookup bound to the requesting principal.
Attack Vector
Exploitation requires network access to the obot API and an authenticated account of any privilege level. The attacker enumerates or obtains an MCP server ID, issues a request to /mcp-connect, and interacts with the restricted MCP server. Tool calls then execute against backend systems using OAuth tokens the platform already holds. See the GitHub Security Advisory GHSA-vw82-7fv8-r6gp and the VulnCheck Advisory - Obot Authorization Bypass for additional detail.
No verified public exploit code is available at this time.
Detect Identity Attacks the Moment Authentication Fails
Singularity™ Identity surfaces credential misuse, lateral movement, and privilege escalation in real time, so you can contain the breach before it spreads.
Stop Identity AttacksDetection Methods for CVE-2026-101084
Indicators of Compromise
- Requests to /mcp-connect from user accounts that were not previously granted access to the targeted MCP server
- MCP tool invocations that originate from unexpected user principals but use OAuth credentials tied to another owner
- Enumeration patterns against MCP server identifiers in application logs
Detection Strategies
- Correlate /mcp-connect request logs against the configured Access Control Rules and alert on mismatches between caller identity and server authorization
- Baseline per-user MCP server usage and flag first-time connections to high-sensitivity servers
- Review audit trails for MCP tool calls executed outside of normal business workflows or by service accounts not previously associated with those tools
Monitoring Recommendations
- Enable verbose audit logging for all /mcp-connect activity, including authenticated user, server ID, and connection outcome
- Monitor downstream systems reached via MCP for anomalous API calls that use the platform's stored OAuth tokens
- Forward obot application and access logs to a centralized analytics pipeline for cross-user correlation
How to Mitigate CVE-2026-101084
Immediate Actions Required
- Upgrade obot to version v0.21.1 or later
- Rotate OAuth credentials stored in obot for MCP servers that may have been reached by unauthorized users
- Review audit logs for prior /mcp-connect requests and MCP tool calls that violate configured Access Control Rules
- Restrict account creation and tighten authentication on obot deployments until patching is complete
Patch Information
The obot maintainers address CVE-2026-101084 in version v0.21.1 by enforcing Access Control Rules on the /mcp-connect endpoint. Patch details are published in the GitHub Security Advisory GHSA-vw82-7fv8-r6gp.
Workarounds
- Place the obot instance behind a reverse proxy that restricts /mcp-connect to a limited allowlist of trusted users until the upgrade is applied
- Temporarily disable MCP servers that hold high-value stored OAuth credentials
- Remove or scope down OAuth tokens associated with sensitive backend systems to reduce blast radius of unauthorized tool calls
# Example: block /mcp-connect at an upstream proxy pending patch deployment
# nginx snippet
location /mcp-connect {
allow 10.0.0.0/24; # trusted admin subnet
deny all;
proxy_pass http://obot_backend;
}
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

