AIBound

Home / Glossary / OWASP MCP Top 10 and the Agentic Top 10

Glossary

OWASP MCP Top 10 and the Agentic Top 10

The OWASP MCP Top 10 and the OWASP Top 10 for Agentic Applications are two OWASP lists of the most serious security risks in MCP servers and in AI agents. The MCP Top 10 is in beta, so its items may still change. The Agentic Top 10 for 2026 was published on 9 December 2025.

By the AIBound team · Updated 17 November 2026

Key takeaways

  • The OWASP MCP Top 10 is in its beta release and pilot testing phase, with items labelled 2025.
  • The OWASP Top 10 for Agentic Applications for 2026 was published on 9 December 2025.
  • The two lists overlap, because agents often reach their tools through MCP servers.

Why these two OWASP lists matter

AI agents plan and take actions, and MCP servers are one common way they reach tools and data. Both bring risks that older lists were not written for. A shared list gives security teams, developers and vendors the same names for those risks.

Employees now install agents and MCP servers on their own laptops. A team that knows the items on both lists knows what to look for when it finds one.

What is on each list and how to use them

Read both lists, then check them against the agents and MCP servers in your company. The item names below are OWASP's own.

  1. 1 Read the MCP Top 10 (beta): Token Mismanagement and Secret Exposure, Privilege Escalation via Scope Creep, Tool Poisoning, Software Supply Chain Attacks and Dependency Tampering, Command Injection and Execution, Prompt Injection via Contextual Payloads, Insufficient Authentication and Authorization, Lack of Audit and Telemetry, Shadow MCP Servers, and Context Injection and Over-Sharing.
  2. 2 Read the Agentic Top 10 for 2026: Agent Goal Hijack, Tool Misuse, Identity and Privilege Abuse, Agentic Supply Chain Vulnerabilities, Unexpected Code Execution, Memory and Context Poisoning, Insecure Inter-Agent Communication, Cascading Failures, Human-Agent Trust Exploitation and Rogue Agents.
  3. 3 Note which items apply to the agents and MCP servers your employees run.
  4. 4 Set a control for each one, and check again when OWASP updates either list.

OWASP MCP Top 10 vs the Agentic Top 10

The MCP list looks at one protocol and the servers that speak it. The agentic list looks at the agent as a whole. They are separate OWASP projects.

OWASP MCP Top 10 vs the Agentic Top 10
Question OWASP MCP Top 10 OWASP Top 10 for Agentic Applications
What does it cover? Risks in MCP servers and how they are used Risks in agentic AI applications as a whole
What is its status? Beta release and pilot testing Published on 9 December 2025 as the list for 2026
How are items numbered? MCP01:2025 to MCP10:2025 ASI01 to ASI10

Sources

Sources checked 17 November 2026

Examples

Examples from the two lists

FAQ

Frequently asked questions

What are the OWASP MCP Top 10 and the Agentic Top 10?

They are two OWASP lists. The MCP Top 10 ranks the main security risks in MCP servers and is in beta. The OWASP Top 10 for Agentic Applications for 2026 ranks the main risks in agentic AI and was published on 9 December 2025. Security teams read them together.

Is the OWASP MCP Top 10 final?

No. On 17 November 2026 the OWASP project page says the project is in its beta release and pilot testing phase. The ten items carry a 2025 label and can still change before a final release. Treat the list as a working reference and check the project page for updates.

What is the difference between the Agentic Top 10 and the LLM Top 10?

The OWASP Top 10 for LLM Applications covers applications built on large language models, and its current version is the 2025 list. The Agentic Top 10 covers AI systems that plan and act with autonomy. A team that runs AI agents usually needs both, since an agent is built on a model.

Which list should a security team start with?

Start with the one that matches what employees run. If MCP servers are in use, read the MCP Top 10 and remember it is in beta. If agents take actions on their own, read the Agentic Top 10. Most companies with coding agents will find both lists apply.

Stop the risky AI your employees already run.

AIBound blocks the high-risk AI on every employee device, in real time. The 14-day trial needs no proof of concept.

Book a Demo