MCPBouncer is a desktop action firewall & live packet inspector for AI IDEs (Cursor & Claude) using MCP. It intercepts local AI tool calls to prevent data destruction and credential leaks: ⢠š Live Inspector: Desktop UI showing every tool call in real time. ⢠šØ Human Approval: Destructive commands (rm -rf, DROP TABLE) pause for instant [ALLOW/BLOCK]. ⢠š DLP: Automatically scrubs OpenAI, AWS & Stripe keys. ⢠š Local SHA-256 tamper-evident audit trail. ⢠100% offline & private.
Hi Product Hunt! š
I built MCPBouncer because giving autonomous AI coding assistants (Cursor, Claude Desktop) direct execution access to local databases, terminals, and file systems creates a massive security blindspot.
One hallucinated `rm -rf /` or `DROP TABLE`, or an API key leaked via prompt injection, can be devastating.
MCPBouncer solves this with a zero-dependency, local desktop firewall:
⢠š Real-time packet inspector dashboard (`http://127.0.0.1:4114`)
⢠šØ Instant Human-in-the-Loop approval for destructive operations
⢠š Automatic DLP secret scrubbing (OpenAI, AWS, GitHub, Stripe)
⢠š¦ Prompt injection & Unicode steganography shield
⢠š Immutable SHA-256 local audit ledger
⢠100% offline & private
I'd love to get your feedback and answer any questions!