Chosen, not claimed.

How a panel works

A review is a panel of three, one per seat. Each seat scores independently; the review is published only when all three have signed off, and it names the server version it applies to. A major release makes the review due again.

• Seat 1

  Insurance practitioner

  Underwriters, brokers, adjusters, agency and carrier operators. They judge whether a server fits how the work is actually done.

• Seat 2

  Compliance and security

  Privacy officers, compliance leads, security engineers. They judge what data leaves, where it goes, and how it is protected.

• Seat 3

  AI and MCP engineer

  People who build and run MCP servers and agents. They judge the tools, the errors, the versioning and the edges.

The review standard

Four criteria, each scored 1–5 by every seat. The scorecard is public, the version is fixed, and the standard is the same for a solo builder and a carrier.

1. 1

   Workflow fit

   Does it do the job an insurance professional actually has, in the order they do it?

2. 2

   Data safety

   What leaves the building, where it goes, and whether PII and policy data are handled the way a regulator expects.

3. 3

   Security

   Authentication, permissions, prompt-injection resistance, and what a compromised client could do.

4. 4

   Build quality

   Tool descriptions, error handling, versioning and whether it behaves the same way twice.