Analyst criteria are written at scoping time and then fixed, so fast-moving areas such as AI agents and MCP servers may be thin or absent in a report. Check whether the criteria name them, then test the gaps yourself against the published MCP authorization rules.
Why new areas lag
Forrester's public method says the analyst sets scope and criteria, then scores vendors against scale explanations written for each criterion. Evaluation takes time, including questionnaires, briefings and a review period. A capability that became important after scoping will not be on the list, whatever its importance to you.
This is not a criticism. It is how a periodic report works. It means you should treat a report as a snapshot of what the analyst considered important at a point in the past.
How to check coverage in minutes
Read the criteria list, not the placement. Search it for:
- Agent identity, meaning how an AI agent is identified and authorized.
- MCP servers, meaning whether tools exposed to agents are inventoried and controlled.
- Guardrails on what agents can call or send.
- Auditability of agent actions.
If none appear, the report was not written to answer your question, and a placement says nothing about it.
What you can test yourself
The MCP specification gives you concrete behavior to check. Its authorization section says, among other things, that MCP servers must validate that access tokens were issued specifically for them, that tokens must not be passed through to upstream APIs, that clients must use PKCE, and that access tokens must not appear in the URI query string. Authorization is optional in the protocol, so ask which servers in your inventory use it.
Turn that into test questions for any vendor claiming MCP or agent security:
- Can the product list the MCP servers in use across your developers and pipelines?
- Can it show which agent or identity called which tool?
- Can it enforce a rule, such as blocking a call, rather than only alerting?
- What does it log, and for how long?
- How does it behave when a server accepts tokens meant for another resource?
Be careful with maturity
New categories carry many early claims. Ask each vendor to label every capability as generally available, early access or roadmap, and write the answer into the contract file or order form. Buy on what is generally available today.
Try it yourself
Open the demo and follow how an MCP server is set up. Notice where an approval or a guardrail would sit in the path.
Click through it at your own pace, or open it full screen.
Safeguard offers an MCP Guardrail, agent identity, policies and gates, and asset discovery as generally available capabilities.
Quick check
Does an absent topic mean the vendors are weak at it?
No. It means the report did not evaluate it. Test the topic directly.
Is MCP authorization required by the protocol?
The specification marks it optional. When supported on HTTP transports, implementations should conform to it.
Sources
Last reviewed: 19 October 2026.
