Skip to main content

Trust Score

LightNow summarizes available trust evidence for an evaluated MCP server as a 0–100 score, a level and a set of evidence layers. Use the result to route a review, not as an automatic approval or guarantee of safe behavior.

Read the result in order

1
Start with the returned level

excellent, high, medium, low, unverified and unknown are routing signals for the review.

2
Check critical failures

A failed_critical layer overrides the aggregate and forces an unverified result.

3
Inspect the evidence layers

Find out whether the weak signal concerns identity, ownership, endpoint, source, reputation or vulnerabilities.

4
Check freshness and intended use

Stale evidence and a high-impact workflow require more review than a disposable personal experiment.

Do not derive the level from the number in client-side policy. Read the level returned by LightNow. This keeps automation aligned when the evaluation contract or thresholds evolve.

Evidence layers

LayerQuestion it helps answer
Publisher identityIs the publisher identity coherent and attributable?
Domain verificationDoes the owner control the claimed domain where that proof applies?
Repository verificationDoes the owner control the referenced source repository?
Endpoint securityIs the remote endpoint reachable through an acceptable TLS and URL posture?
Source codeWhat source, license and provenance evidence is available?
ReputationWhat maturity and activity signals are visible for the upstream project?
VulnerabilityAre known advisories associated with the evaluated source or package?
System blocklistDoes a critical rule require the result to be rejected?

not_applicable means a layer has no relevant input for this server. It is excluded from the weighted aggregate instead of lowering the score. A missing or failed applicable layer is different and should remain visible in the review.

How aggregation works

Several layers can have configured weights. LightNow normalizes the weighted sum across the layers that apply. This means the visible percentages are not a simple checklist that must add up to 100 before normalization.

The numeric score is useful for comparison. The layer status and evidence are what explain the decision.

Interpret the levels

LevelInterpretation
excellent or highNo critical failure is present and the available evidence is strong. Still confirm capabilities and data access for the intended workflow.
medium or lowAt least one relevant signal is weaker or incomplete. Inspect the layer and decide whether isolation or additional proof is sufficient.
unverifiedA critical failure or explicit unverified outcome requires investigation before rollout.
unknownThe evaluation has not produced enough applicable data. Unknown is not the same as safe.

Trust is contextual

The same server can be acceptable for a temporary personal test and unsuitable for an organization-wide default profile. Consider:

  • which tools, resources and prompts it exposes,
  • which local files or remote systems it can access,
  • whether the server handles production or sensitive data,
  • whether ownership and source evidence match your policy, and
  • whether the evidence is fresh enough for the rollout decision.

Trust Score does not replace sandboxing, least privilege, secret scoping or an internal security review.

When to re-check

Review trust again when the server version, publisher, source repository, endpoint or vulnerability posture changes. Also re-check before promoting a server from a personal profile into an organization default.