cpex-secrets-detection 0.1.0

High-performance secrets detection and redaction for MCP Gateway

Secrets Detection (Rust)

High-performance secrets detection and redaction for ContextForge and MCP Gateway.

Features

• Rust-owned recursive scanning for strings, dicts, lists, and MCP payload containers
• Built-in detection for high-signal credential formats such as AWS access keys and Slack tokens
• Optional redaction or hard blocking on detection
• Hook coverage for prompt input, tool output, and fetched resource content
• Opt-in broad patterns for lower-confidence generic token assignments
• Structured findings metadata with redacted match previews

Build

make install

Usage

The plugin scans data at these hook points:

• prompt_pre_fetch
• tool_post_invoke
• resource_post_fetch

Typical uses:

• redact secrets before they leave the gateway
• block tool or resource payloads that contain leaked credentials
• add lightweight findings metadata for downstream auditing

Configuration

Key settings include:

• redact: replace detected secret values in the payload
• redaction_text: replacement string for detected values
• block_on_detection: stop processing instead of modifying payloads
• min_findings_to_block: threshold before hard blocking
• enabled: enable optional lower-confidence detectors when needed

Detection Notes

• High-signal built-in patterns are enabled by default.
• Broader generic assignment-style patterns are opt-in to avoid noisy false positives.
• Findings include secret type labels such as aws_access_key_id or slack_token.
• Redaction preserves surrounding structure when possible, for example replacing only the secret value inside a larger assignment string.

Returned Metadata

When detections occur, the plugin can return metadata such as:

• count
• secrets_redacted

Blocking responses use the SECRETS_DETECTED violation code.

Testing

make ci

Security Notes

• Default detectors are intentionally biased toward high-confidence secret formats.
• Broad token-like patterns should only be enabled when your environment benefits from higher recall and can tolerate extra review.