Skip to content

Idea: Runtime trust scoring based on actual execution history? #1

@xkumakichi

Description

@xkumakichi

Hey — really interesting approach with static CI scanning for MCP servers. I've been thinking about the same problem from a different angle.

Your approach: Scan the server code/config for risk signals before deployment.
My question: What about runtime behavior? A server might pass all static checks but still fail 20% of the time in production.

I've been experimenting with a lightweight middleware that logs actual tool executions (success/fail, latency) and gives a simple trust verdict based on real history:

can_I_trust("my-server") → { verdict: "yes", success_rate: 99.2% }

It feels like static analysis (what you're doing) + runtime verification could be really complementary.

Curious:

  1. Have you thought about incorporating runtime data into trust scoring?
  2. Would a "runtime trust score" that feeds into your risk assessment be useful?

Here's my experiment if you want to look: https://github.com/xkumakichi/veridict

No pressure at all — just exploring this space and your project caught my eye.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions