Stop Confidential Leaks from AI Chat History! The Mac Exclusive Scanner “Sieve” is Revolutionary!

📰 News Overview

• The Mac app “Sieve” has been released to detect leaked API keys and passwords from AI chat histories (Cursor, Claude Code, VS Code Copilot, etc.).
• Runs 100% locally, utilizing macOS’s Keychain to securely store and manage sensitive information without any network communication.
• Features the ability to directly redact (remove) confidential information from the SQLite database of VS Code.

💡 Key Points

• Extensive Scan Targets: Covers major AI tool directories like ~/.claude/, Application Support/Cursor/, and even .env files within projects.
• MCP Integration: Comes with a local MCP server for Claude Code, allowing commands to be executed using credentials stored in the Vault without exposing raw keys to Claude.
• Privacy-First Design: No account required, no telemetry, and based on an open-source core (SieveCore), tailored for security-conscious developers.

🦈 Shark’s Perspective (Curator’s View)

In 2026, where letting AI agents write code is the norm, accidents happen where live API keys slip into prompts! What’s amazing about “Sieve” is that it doesn’t just “find” but can directly “erase” information from the VS Code chat database. Its implementation of handling SQLite while taking backups is super specific and shows a deep understanding of developer pain points! Plus, since it operates as an MCP server, you can get Claude to do its thing without exposing raw keys. For any security-conscious shark, it’s definitely worth the $9.99 investment!

🚀 What’s Next?

As AI coding assistants become more widespread, tools like this are bound to become “standard gear” in development environments. There’s a chance they’ll be integrated directly into IDEs in the future, but for now, protecting yourself with robust standalone tools like Sieve is the way to go!

💬 A Word from Haru-Same

Chat histories are like a shark’s stomach; you never know what’s lurking inside! Better clean up while you can!

📚 Terminology Explained

• MCP (Model Context Protocol): A standard for AI models to interact safely with external tools and data.

• Redaction: Refers to editing or deleting. In this tool, it pertains to the function of concealing or removing confidential information within the database.

• Security-scoped bookmarks: An Apple security mechanism allowing sandboxed apps on Mac to continuously access specific folders with user permission.

• Source: Sieve – scans Cursor/Claude chat history for leaked API keys