CVE-2026-61427: MervinPraison PraisonAI authentication bypass in MCP HTTP-stream transport

CVE-2026-61427 · Severity: high · CVSS 7.3 · Published 2026-07-15

PraisonAI, a framework for managing AI agents, contains a security flaw where its Model Context Protocol (MCP) server is accessible without a password by default. If an operator runs the server on a public network address, an unauthorized user could connect to the system, view available AI tools, and trigger tool functions. This could lead to unauthorized data access or abuse of AI services and associated API costs.

Technologies: praisonai (PyPI). Vendors: PyPI.

Quick answers

Is CVE-2026-61427 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-61427 being exploited in the wild.
How severe is CVE-2026-61427?
High severity, with a CVSS score of 7.3, published 15 July 2026.
Which products and versions are affected?
praisonai (PyPI) <= 4.6.77, ECOSYSTEM:introduced=0,fixed=4.6.78.
Is there a fix for CVE-2026-61427?
Yes. praisonai (PyPI): upgrade to 4.6.78 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-61427 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-61427?
Check whether you run praisonai (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • praisonai (PyPI): upgrade to 4.6.78 or later
    pip install "praisonai>=4.6.78"

Technical details

PraisonAI's MCP HTTP-stream transport (src/praisonai/praisonai/mcp_server/transports/http_stream.py) only enforces Authorization/Bearer checks if an API key is configured; however, the CLI --api-key option defaults to None. Consequently, an unauthenticated attacker can initialize a session and access the tools/list and tools/call endpoints. Furthermore, the dispatcher in mcp_server/server.py forwards tool-call arguments directly to handlers without validating them against the advertised inputSchema, allowing for the injection of undeclared parameters. While the server binds to 127.0.0.1 by default, remote exploitation is possible if the operator binds to a network-accessible address (e.g., via --host 0.0.0.0). The issue is resolved in version 4.6.78.

Affected products

  • MervinPraison PraisonAI < 4.6.78

Identifiers

  • CVE-2026-61427
  • GHSA-hc5v-gxvj-58wh
  • PYSEC-2026-4257

Timeline

  • 2026-06-25: advisory: GitHub Security Advisory published
  • 2026-07-15: disclosed: NVD publication date
  • 2026-07-15: patched: Fix confirmed in version 4.6.78

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.