Typosquatted Blender MCP Repo Hit 958 Stars in Days and Silently Ran Malware
Promptmethus · x · 2026-10-23
A typosquatted Blender AI MCP repo was exposed: the fake LiCinCan/blender-ai-mcp-addon differs by one letter from the authentic ahujasid/blender-ai-mcp-addon and even carries the original author's name in its license header. The account registered on Oct 15 and racked up 958 stars and 388 forks within days; its issues were copied from the real repo.
Malware mechanics: once the plugin is enabled, it downloads a Python environment from another repo under the same account, extracts hidden code from a .blend file, and silently executes it outside Blender while swallowing all errors. The 114MB release contains a segmented exe stitched together—classic malware delivery.
Promptmethus, sharing the warning, says this is why he's rewriting everything down to the OS and his own language in C++. Lesson: have AI inspect the source of any MCP/Skill/plugin before installing, regardless of the promoter's follower count.
More from coding & agent
- Devs on Reddit swap tactics for stopping prompt regressions in AI agents — SleepFlimsy6963 · 2026-10-23
- Claude Code + ComfyUI MCP workflow for client work: paid models or open weights? — newaccount47 · 2026-10-23
- Social scientist: CS treats agentic development as a magic black box — RexDouglass · 2026-10-23
- Someone vibe-coded a Tesla Roadster game — before GTA VI — Jasonio · 2026-10-23
- Codex 0.164.0 'account/read failed' TUI error — fix is restarting the app-server daemon — bytebot · 2026-10-23
- Same native UI, two stacks: Swift build is ~80MB Mac-only, Zig build is ~10MB cross-platform — jedisct1 · 2026-10-23