Executive brief
PraisonAI is a multi-agent system that orchestrates AI tasks through tool execution. Legacy HTTP endpoints in versions before 1.6.59 (agents) and 4.6.59 (main product) lack proper authentication and cross-origin controls, allowing a malicious website to use DNS rebinding techniques to invoke tools on the server with the privileges of the user running the service. This could enable unauthorized command execution, data exfiltration, or other harmful operations depending on available tools.
Technical details
The vulnerability is a cross-origin request forgery (CORS bypass) combined with weak access controls in the ToolsMCPServer.run_sse() method in src/praisonai-agents/praisonaiagents/mcp/mcp_server.py. The legacy /sse and /messages/ endpoints mount SseServerTransport without enforcing Host, Origin, or authentication headers. An attacker on a malicious website can exploit DNS rebinding to redirect the target's browser to a local or internal SSE server, supply arbitrary Host and Origin headers, enumerate registered tools via the HTTP interface, and invoke them with the server's user privileges. The newer Streamable HTTP transport correctly rejects hostile origins, confirming the flaw is specific to the legacy SSE wrapper. Patches were released in praisonaiagents 1.6.59 and PraisonAI 4.6.59.
Affected products
- PraisonAI PraisonAI 3.10.0 to 4.6.58
- PraisonAI praisonaiagents 0.6.0 to 1.6.58
Timeline
- 2026-09-15: disclosed: Published on NVD
- 2026-06-13: patched: Patch committed (input validation and access control hardening)
- 2026-06-17: patched: PraisonAI v4.6.59 released