Klarion

•
11
•

Detects leaked API keys, tokens, and credentials in code and git history using entropy detection and LLM-powered adjudication.

Klarion is an AI-powered secrets scanner designed to eliminate false positives in credential detection by pairing fast local heuristic checks with intelligent LLM adjudication. Operating directly in developer workflows—including CLI, pre-commit hooks, GitHub Actions, MCP server mode, and Claude Code hooks—it detects candidate secrets via entropy and pattern matching before asking an AI model to evaluate contextual code. This two-stage pipeline ensures test fixtures, hashes, and non-sensitive identifiers are ignored while real credentials from services like AWS, Stripe, and GitHub are reliably caught before they land in version control.

Key Features

01Two-stage pipeline combining local pattern/entropy detection with LLM context adjudication
02Near-zero false positive rate by suppressing test values, documentation examples, and hashes
03Compatible with multiple model backends including Anthropic, OpenAI, local Ollama, or active Claude Code sessions
04Privacy controls with optional secret masking and zero-setup evaluation delegation via MCP
0511 GitHub stars
06Runs natively as an MCP server, CLI, pre-commit hook, Claude Code hook, and GitHub Action

Use Cases

01Scanning pull requests and CI pipelines to block leaked tokens from entering production git branches
02Preventing AI coding agents in Cursor, Cline, or Claude Code from hardcoding API keys and credentials into source files
03Auditing existing repositories and baselining legacy code to surface exposed credentials without alert fatigue