Trust Scoring

Every MCP server deployed through Conduit is scanned for security vulnerabilities, compliance issues, and performance problems. Results are distilled into a trust score and grade used by MLSs to evaluate vendor access requests.

Five scoring dimensions

CategoryWeightWhat it checks
Security40%Tool poisoning, prompt injection, data exfiltration, command injection, path traversal, credential exposure
Compliance20%MCP spec adherence, error handling, transport configuration, protocol version
Auth20%Authentication mechanisms, TLS configuration, CORS headers, credential storage
Performance10%Response latency, error rate, connection stability, throughput
Runtime10%Gateway telemetry analysis (requires active Conduit deployment)

Each category scores 0-100. The total score is the weighted sum, rounded to the nearest integer.

Grade scale

A+

95-100

A

90-94

B+

85-89

B

80-84

C+

75-79

C

70-74

D

60-69

F

< 60

13 finding categories

tool_poisoningTool descriptions contain hidden instructions or manipulation attempts
prompt_injectionTool outputs or resource contents contain injected prompts
data_exfiltrationTools attempt to send data to unauthorized external endpoints
command_injectionTool parameters allow arbitrary command execution
path_traversalFile operations allow access outside intended directories
credential_exposureSecrets or credentials exposed in tool definitions or responses
auth_weaknessMissing or insufficient authentication/authorization
spec_violationNon-compliance with MCP protocol specification
transport_issueTLS, CORS, or transport configuration problems
performance_issueHigh latency, timeout, or error rate
runtime_anomalyUnusual patterns detected in gateway telemetry
hidden_contentHidden or obfuscated content in tool metadata
cross_server_riskInteractions that could affect other servers or clients

Severity levels

[!!]
criticalImmediate exploitation risk. Server should not be used until resolved.
[! ]
highSignificant security concern. Should be addressed promptly.
[~ ]
mediumModerate risk. Should be addressed in the next update.
[. ]
lowMinor concern or best practice recommendation.
[i ]
infoInformational. No action required.

Scan process

Scans are triggered on deploy, on schedule (based on plan), or manually:

bash
# Via API
POST /api/servers/:slug/scan

# Via CLI
conduit scan my-server

Scans typically take 10-30 seconds. The scanner connects to your server, discovers tools, analyzes tool definitions and behaviors, and produces findings with scores.

Badge embeds

Display your trust score with embeddable SVG badges in three styles:

Standard

<img src="https://conduitapi.dev/api/badges/{serverId}" alt="Trust Score" />

Compact

<img src="https://conduitapi.dev/api/badges/{serverId}?style=compact" alt="Trust Grade" />

Markdown (for READMEs)

[![Conduit Trust](https://conduitapi.dev/api/badges/{serverId})](https://conduitapi.dev/trust/{serverId})