Hey Product Hunt! As the Model Context Protocol (MCP) rapidly becomes the standard for connecting LLMs to databases, terminals, and internal APIs, a silent security challenge has emerged: **Shadow MCP Servers**. Developers and teams are configuring MCP servers across multiple AI tools (Claude Desktop, Cursor, VS Code, Gemini IDE) or exposing local HTTP servers on standard ports often completely unauthenticated, with access to internal company data and local shells. We built MCPScan to give developers and SecOps teams complete visibility into their MCP attack surface. ### What MCPScan does: - **Dual-Transport Discovery:** Scans private CIDR networks for HTTP MCP servers and inspects local AI tool configurations (Claude Desktop, Cursor, Antigravity, VS Code). - **3-Layer MCP Verification:** Differentiates real MCP servers from generic JSON-RPC gateways and Ethereum nodes to prevent false positives. - **Single-Probe Auth Auditing:** Sends exactly 1 unauthenticated request per server to check whether tools/resources are exposed without authorization (no brute forcing, zero state changes). - **Active vs. Dormant Process Matching:** Cross-references local configs with running OS processes to find active servers and dormant shadow configurations. - **Privacy & Secret Security:** 100% offline, zero telemetry, zero storage of `env` variable contents, and automatic masking of sensitive API keys/tokens in CLI arguments. It's compiled as a single zero-CGO binary for Windows, macOS, and Linux with embedded SQLite reporting. Would love to hear your thoughts, feedback, and any other AI tools you'd like us to support next!

MCPScan

Discover & audit shadow Model Context Protocol servers

Discover & audit shadow Model Context Protocol servers