/

CVE-2026-58197

Severity:

HIGH RISK

(8.8)

Improper Access Control (CWE-284) in ToolHive MCP Servers

CVE-2026-58197

Deferred

Vulnerability type

Category:

Improper Access Control

CWE:

CWE-284, CWE-306

Impact:

Unauthorized access and lateral movement

Severity:

HIGH RISK

(8.8)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

ToolHive MCP server containers may run without network isolation, allowing a malicious or compromised server to reach host-local services, ToolHive proxies, and control-plane endpoints through the Docker gateway. Because these endpoints may be accessible without authentication, an attacker could access data, invoke sibling tools, modify workload state, or disrupt services without escaping the container.

Affected context

The issue affects ToolHive deployments where locally run MCP server containers use the default network permission profile instead of an isolated network policy. It is especially relevant when the ToolHive API, MCP proxy endpoints, or control-plane services are reachable from the container network without authentication. ToolHive Studio configurations that explicitly disable network isolation can increase exposure.

Exploit preconditions

  • Attack surface: Network connectivity from an MCP server container to Docker gateway and ToolHive-managed endpoints
  • Required access: Ability to deploy, control, or compromise an MCP server used by ToolHive
  • Trigger point: Network requests made by the MCP server to host-local services, sibling proxies, or the ToolHive control plane
  • Why exploitation is possible: Default or explicitly disabled network isolation permits connectivity, while reachable endpoints may lack authentication and authorization controls

Why this happens

The vulnerability results from an insecure network trust boundary around locally run MCP server containers. Containers are granted connectivity to host-local and ToolHive-managed services without enforcing a sufficiently restrictive network policy, and some reachable endpoints do not require authentication.

This maps primarily to CWE-284: Improper Access Control, because access to protected services is not adequately restricted based on the container's identity or role. It also relates to CWE-306: Missing Authentication for Critical Function, where sensitive API or proxy operations may be reachable without verifying the caller.

Potential impact

  • An attacker may access host-local services and exposed operational data
  • A compromised MCP server could invoke tools provided by sibling MCP servers
  • Attackers could read logs, configuration data, or other sensitive information
  • Unauthorized requests could alter process or workload state
  • Service availability could be disrupted
  • The attacker could move laterally across ToolHive-managed components without escaping the container

General mitigation guidance

Mitigation generally focuses on reducing container network trust and requiring authenticated, authorized access to internal services.

Common defensive patterns include:

  • Enable network isolation by default for locally run MCP containers
  • Use explicit allowlists for required destinations and protocols
  • Avoid exposing host-local services through broad Docker gateway connectivity
  • Require authentication and authorization for ToolHive APIs, MCP proxies, and control-plane operations
  • Apply least-privilege identities and narrowly scoped service permissions
  • Treat MCP server containers as potentially untrusted workloads
  • Prevent client or UI settings from silently overriding secure backend isolation defaults
  • Monitor and restrict access to sensitive logs, workload controls, and administrative endpoints

How teams detect issues like this

Organizations can detect this class of issue through:

  • Static analysis and configuration checks for disabled network isolation or permissive container network profiles
  • Code review searches for boolean settings that override secure isolation defaults
  • Inventory and policy checks identifying unauthenticated API, proxy, or control-plane endpoints
  • Network testing from MCP containers to host-local addresses, Docker gateway services, and sibling proxies
  • Runtime monitoring for unexpected container-to-host or container-to-container connections
  • Authorization tests confirming that sensitive operations reject unauthenticated and unauthorized callers
  • Automated security review tools that flag missing authentication and overly broad network permissions

Sources

SourceLink
github.comgithub.com/stacklok/toolhive-studio/commit/968182d7f3ee1e55123369e66ad88f82128119b0
github.comgithub.com/stacklok/toolhive-studio/pull/2469
github.comgithub.com/stacklok/toolhive-studio/releases/tag/v0.38.0
github.comgithub.com/stacklok/toolhive/commit/d8f40cb1599b8bf66657f2dfff15bfbfc236e712
github.comgithub.com/stacklok/toolhive/pull/5583
github.comgithub.com/stacklok/toolhive/releases/tag/v0.30.1
github.comgithub.com/stacklok/toolhive/security/advisories/GHSA-qg2g-g9w3-m5h8

Frequently asked

questions

Exploitation depends on the deployment's network exposure and the attacker's ability to control or compromise an MCP server. A malicious server can use its permitted network position to reach exposed internal services.

No host access or container escape is necessarily required. The vulnerable network permissions may allow a compromised MCP server to contact reachable host-local and ToolHive-managed endpoints from inside its container.

Enforce network isolation by default, restrict permitted destinations, and require authentication and authorization for all sensitive internal APIs and proxy operations.

Treat MCP servers as untrusted workloads, apply least-privilege networking and service permissions, prevent insecure configuration overrides, and continuously test container-to-service access boundaries.

Find out what's already
exploitable in your codebase.