highAgent ThreatProtocol Vulnerability

PraisonAI MCP Server Missing Default Authentication (Unauthenticated Tool Access)

First seen Jul 15, 2026 · Updated Jul 15, 2026 · CVSS 7.3

MCPunauthenticated-accesstool-calldefault-configmissing-authinput-validationpraisonaiASI06 · Memory PoisoningSurface: ProtocolPropagation: Single Hop

PraisonAI versions before 4.6.78 default to running the MCP HTTP-stream server without any API key or authentication, meaning anyone who can reach the endpoint can list and invoke all exposed tools. The server also fails to validate tool-call arguments against the advertised schema, compounding the risk of malformed or malicious inputs reaching tool handlers. Exploitation requires the operator to have bound the server to a network-accessible address rather than the safe localhost default.

Technical Analysis

The PraisonAI CLI's 'mcp serve --transport http-stream' command only enforces Authorization/Bearer header checks when an --api-key is explicitly supplied; the default is None, leaving the endpoint open. Because Origin header checks are also not enforced, an unauthenticated remote or local network client can initialize an MCP session, call tools/list to enumerate capabilities, and tools/call to invoke arbitrary registered tools. The dispatcher further forwards call arguments directly to tool handlers without validating them against the declared inputSchema, so downstream tool logic may receive unexpected or malicious parameter shapes, increasing the chance of injection, path traversal, or logic-abuse bugs in tool implementations. The blast radius is limited by the safe 127.0.0.1 bind default, but any operator who binds to 0.0.0.0 (common in containerized or cloud deployments) exposes the full tool surface to unauthenticated network attackers, crossing the trust boundary between the MCP protocol layer and the underlying agent/tool execution layer.

Affected Systems

PraisonAI; protocols: MCP

Detection Signatures

  • MCP requests to /mcp or http-stream endpoints lacking an Authorization header
  • Absence of Origin header validation in server logs
  • tools/list or tools/call requests from unauthenticated/unexpected source IPs
  • PraisonAI process launched with --transport http-stream and --host 0.0.0.0 without --api-key
  • Tool-call payloads with fields not matching the tool's inputSchema

Remediation Steps

  1. 1

    Upgrade PraisonAI

    Update to PraisonAI 4.6.78 or later where authentication defaults and schema validation are fixed.

  2. 2

    Always set an API key

    Explicitly configure --api-key (or equivalent auth mechanism) whenever running the MCP HTTP-stream server, even in test/dev environments.

  3. 3

    Restrict network binding

    Keep the server bound to 127.0.0.1 unless remote access is required; if remote access is needed, place it behind a reverse proxy or VPN with its own authentication.

  4. 4

    Validate tool inputs independently

    Add server-side or tool-level validation of arguments against inputSchema regardless of dispatcher behavior, to prevent malformed calls from reaching handler logic.

  5. 5

    Enforce Origin/CORS checks

    Add Origin header validation and CORS restrictions at the proxy or server layer to reduce cross-origin abuse even if API key is later enabled.

  6. 6

    Monitor MCP access logs

    Alert on tools/list and tools/call requests without valid Authorization headers or from unexpected network ranges.

CVE / Advisory IDs

CVE-2026-61427

Industries Most Exposed

Software/TechnologyAI/ML platform providersCloud servicesAny organization self-hosting PraisonAI agents

Sources

Respond to this threat

Pro subscribers get a full AI-generated incident-response playbook for this threat — detection, containment, eradication, and recovery steps — plus an unlimited AI Threat Advisor for questions about your environment.