CVE-2026-57123: PraisonAI: MCP SSE transport binds 0.0.0.0 with no authentication and no Origin validation; bundled SecurityConfig is never wired in
The MCP SSE server started via ToolsMCPServer.runsse() / launchtoolsmcpserver(transport="sse") binds to 0.0.0.0 by default and builds its Starlette application with no authentication middleware and no Origin-header validation.
Published Jun 18, 2026 · Updated Jul 20, 2026
Affected packages
praisonaiagents
- Ecosystem
- PyPI
- Affected
- < 1.6.59
- Fixed in
- 1.6.59
- Check
pip show praisonaiagents
Advisory details
From the GitHub Advisory Database (CC BY 4.0).
The MCP SSE server started via ToolsMCPServer.run_sse() / launch_tools_mcp_server(transport="sse") binds to 0.0.0.0 by default and builds its Starlette application with no authentication middleware and no Origin-header validation. The module mcp/mcp_security.py provides exactly the needed controls (origin validation, DNS-rebinding detection, auth-header enforcement, a SecurityConfig), but none of these functions are ever called by any transport — they are dead code. Any host that can reach the port can list and invoke every registered tool with no credentials, and a victim's browser can drive the same calls against a localhost instance via DNS rebinding.
Affected code: src/praisonai-agents/praisonaiagents/mcp/mcp_server.py
- run_sse defaults host to all interfaces (line 245) and builds the app with only
debugandroutes - no
middleware=and no per-route auth/origin gate (lines ~271-289): app = Starlette(debug=self._debug, routes=[ Route(sse_path, endpoint=handle_sse), # "/sse" Mount(messages_path, app=sse_transport.handle_post_message), # "/messages/" ]) uvicorn.run(app, host=host, port=port) - launch_tools_mcp_server also defaults host="0.0.0.0" (line 301).
src/praisonai-agents/praisonaiagents/mcp/mcp_security.py defines but the transports never call:
- is_valid_origin (line 30), is_potential_dns_rebinding (line 110), validate_auth_header (line 167), SecurityConfig.is_origin_allowed (line 236). These symbols are referenced only inside mcp_security.py and the __init__ re-export. (mcp_websocket.py's auth references are CLIENT-side, not server validation.)
Impact: launch_tools_mcp_server(transport="sse") is the documented path for exposing tools over MCP. With the defaults above it is an unauthenticated, network-reachable tool-execution endpoint. Blast radius equals the capabilities of the registered tools; with file/shell/code-exec tools this is RCE. With no Origin check, a malicious page the victim merely visits can rebind its hostname to 127.0.0.1 and issue the JSON-RPC calls cross-origin against a developer's local server.
Proof of concept: Static proof (AST analysis of unmodified source): Check 1 - run_sse(host='0.0.0.0'); launch_tools_mcp_server(host='0.0.0.0') -> EXPOSED Check 2 - Starlette(...) kwargs: ['debug','routes'] -> NO middleware= (no auth/origin gate) Check 3 - is_valid_origin / is_potential_dns_rebinding / validate_auth_header / SecurityConfig never called by any transport -> DEAD CODE Live exploitation against a running server: curl -N http://VICTIM:8080/sse
event: endpoint / data: /messages/?session_id=
curl -X POST "http://VICTIM:8080/messages/?session_id=" -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05", "capabilities":{},"clientInfo":{"name":"x","version":"1"}}}' curl -X POST "http://VICTIM:8080/messages/?session_id=" -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"","arguments":{...}}}' No Authorization header anywhere. Browser DNS-rebinding variant drives the same calls cross-origin.
Remediation: Wire in the existing mcp_security.py controls and fix defaults:
- Default run_sse(host="127.0.0.1"); require explicit opt-in to bind 0.0.0.0.
- Attach Starlette middleware calling is_valid_origin / is_potential_dns_rebinding; reject bad origins.
- Enforce validate_auth_header when SecurityConfig.require_auth; default require_auth=True (and allow_missing_origin=False) for any non-loopback bind.
Distinct from prior advisories: The accepted MCP advisories are tool-handler bugs — tools/call path traversal -> .pth RCE (GHSA-9mqq-jqxf-grvw) and unauthenticated file read via workflow.show/validate (GHSA-9cr9-25q5-8prj). This is a transport-layer missing-auth/exposure: the SSE server never enforces auth or Origin validation and ignores the security module the codebase ships. Closest in spirit to the default-insecure pattern (GHSA-8444 / 86qc) but a different server and a different root cause (unwired controls, not an unset env var).
Credited to sour-exploit.
References
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. No analysis has been written for this entry yet. Methodology
More PyPI vulnerabilities
All PyPI entriesCVE-2026-61447
PraisonAI: CodeAgent Executes LLM-Generated Code Without Sandboxing and Leaks All Environment Secrets
critical10.0Oct 8, 2026CVE-2026-57112
PraisonAI ToolsMCPServer legacy SSE transport accepts attacker Host/Origin and exposes registered tools
high8.3Jun 18, 2026CVE-2026-57118
PraisonAI AgentTeam.launch exposes unauthenticated remote agent listing and invocation endpoints
critical9.8Jun 18, 2026CVE-2026-57125
PraisonAI: Unauthenticated RCE via Jobs API + Approval Bypass
critical9.8Jun 18, 2026CVE-2026-47392
PraisonAI vulnerable to sandbox escape via `print.__self__` builtins module leak in `execute_code` (subprocess mode)
critical9.9May 29, 2026CVE-2026-40111
PraisonAIAgents has an OS Command Injection via shell=True in Memory Hooks Executor (memory/hooks.py)
critical9.3Apr 10, 2026
