cyber.hendryadrian.com
Security News · securityweek.com

Enterprise MCP Specification Shifts Security Risks to Server Implementations

· Original article ↗

Summary

Akamai says the MCP 2026-07-28 specification reduces protocol-level risks but shifts security responsibility to server implementers, who must address potential workflow hijacking, cross-tenant access, header leaks, XSS and DoS.

Key points

  • The MCP 2026-07-28 update changes the protocol to a stateless model and provides a 12-month deprecation window for legacy versions.
  • Akamai says the update addresses session hijacking and unsolicited server-initiated prompts and strengthens authentication.
  • Tracking identifiers and state objects could expose workflows to hijacking, cross-agent data access or unauthorized cross-tenant actions if implementation is weak.
  • New MCP-specific HTTP headers may enable protocol-confusion attacks or expose secrets if sensitive values are mapped into headers.
  • MCP Apps may introduce browser risks such as stored XSS, while long-running tasks could enable resource-exhaustion DoS attacks.
  • The article emphasizes that these are potential implementation risks; security depends increasingly on MCP server developers and platform operators.

Article Details

Event Type
Planned MCP specification update introducing enterprise-scale, cloud-native capabilities and new implementation-dependent security risks
Impact
The MCP 2026-07-28 specification is expected to shift more security responsibility to MCP server developers and platform operators. The article describes potential risks including workflow hijacking, cross-tenant access, secrets leakage through headers, protocol-confusion attacks, denial of service against long-running tasks, and script execution or phishing through insecure UI panels. These are prospective implementation risks, not reported exploitation of the specification.

People

Maxim ZavodchikAkamai senior director of threat research who described how the stateless model, rich UI apps, and asynchronous tasks make security boundaries dependent on implementation.

Vendors

AkamaiAkamai is one of the firms that has studied the new format ahead of the July 28 launch and describes its own conclusions in a blog report. For cybersecurity, “While the protocol removes several classes ofAnthropicIt was introduced by Anthropic in 2024 and has since become the de facto standard for connecting AI agents to business tools.

Related Articles