/

CVE-2026-57139

Severity:

CRITICAL RISK

(9.8)

Unauthenticated MCP Access (CWE-306) in PraisonAI TS

CVE-2026-57139

Deferred

Vulnerability type

Category:

Missing Authentication

CWE:

CWE-306, CWE-862, CWE-1188

Impact:

Unauthorized tool execution and sensitive data disclosure

Severity:

CRITICAL RISK

(9.8)

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

PraisonAI's HTTP MCP server accepts requests without authentication or authorization and forwards them to registered MCP handlers. Any network client that can reach the server port may invoke tools, read resources, or retrieve prompts using the server's credentials and process privileges.

Affected context

The vulnerability occurs in deployments using the TypeScript MCP server's HTTP transport where the server listens on a reachable interface and does not enforce caller identity or permissions. It is particularly risky when the endpoint is exposed beyond a trusted local network or when registered handlers can access sensitive data, external services, or privileged operating-system resources.

Exploit preconditions

  • Attack surface: The MCPServer HTTP endpoint and its request-handling path
  • Required access: Network connectivity to the listening port; valid application credentials are not required by the affected endpoint
  • Trigger point: HTTP POST requests that invoke MCP operations such as tools/call, resources/read, tools/list, or prompts/get
  • Why exploitation is possible: Requests are forwarded to registered handlers without authentication or authorization checks, allowing untrusted clients to use the server's execution context and credentials

Why this happens

The underlying bug is an authorization design failure in an HTTP-facing service. The server binds without an appropriate host restriction and dispatches incoming requests directly to MCP handlers without first establishing the caller's identity or verifying permission for the requested operation.

This maps primarily to CWE-306: Missing Authentication for Critical Function, with related CWE-862: Missing Authorization and CWE-1188: Insecure Default Initialization of Resource concerns. The insecure pattern is exposing a powerful request dispatcher with permissive network reachability and no secure-by-default access control.

Potential impact

  • An unauthenticated client may invoke registered tools
  • Attackers could access or disclose registered resources and prompt data
  • Handlers may perform actions using server-side credentials
  • The application process's privileges could be abused for unauthorized operations
  • Connected services or downstream systems may be modified or queried

General mitigation guidance

Mitigation generally focuses on reducing endpoint exposure and enforcing access control before dispatching MCP requests.

Common defensive patterns include:

  • Binding administrative or local-only services to a restricted interface by default
  • Requiring strong authentication for every HTTP request
  • Enforcing per-tool, resource, and prompt authorization decisions
  • Separating untrusted clients from handlers that hold sensitive credentials
  • Applying least-privilege permissions to the server process and its service accounts
  • Using network segmentation, firewall rules, and authenticated reverse proxies where appropriate
  • Returning only the minimum data needed by authorized callers

How teams detect issues like this

Detection can combine code analysis, configuration review, and runtime monitoring:

  • Use static analysis to identify HTTP handlers that dispatch requests without authentication or authorization checks
  • Review calls to server binding and listener APIs for unrestricted host or interface selection
  • Inspect MCP routes for consistent identity verification and operation-level permission checks
  • Monitor access logs for unexpected clients invoking tools/call, resources/read, or other MCP operations
  • Enforce CI and deployment policies requiring authentication on non-local service endpoints
  • Use automated code review tools to flag externally reachable request dispatchers that lack access-control middleware

Sources

SourceLink
github.comgithub.com/MervinPraison/PraisonAI/commit/709a0387853709a5e2f2c34680fa64cada252c27
github.comgithub.com/MervinPraison/PraisonAI/releases/tag/v4.6.60
github.comgithub.com/MervinPraison/PraisonAI/security/advisories/GHSA-j4f3-55x4-r6q2

Frequently asked

questions

Yes, when the HTTP MCP port is reachable by an attacker. The attacker generally does not need local access if network routing and firewall rules permit connectivity.

Not at the vulnerable endpoint. A client that can reach the service may submit MCP requests without the authentication and authorization checks needed to restrict access.

Keep the service inaccessible to untrusted networks, require authentication before request dispatch, and apply authorization checks for each tool, resource, and prompt operation.

Treat MCP endpoints as privileged APIs, bind them to restricted interfaces by default, enforce access-control policies in code and at the network boundary, and test unauthenticated and unauthorized request paths in CI.

Find out what's already
exploitable in your codebase.