Trapdex Subscribe

CampaignsAI plugin / MCP / skill

SANDWORM_MODE: npm worm that plants rogue MCP servers in AI assistant configs

Do this now

  1. Audit MCP configs for Claude Desktop, Cursor, Continue and Windsurf for unknown server entries
  2. Remove the listed packages and rotate npm, GitHub, CI, cloud and LLM API secrets
  3. Block the exfil and fallback domains

Indicators to check

Look for any of these on your machines or in your logs:

  • packageclaud-code@0.2.1
  • packagecloude@0.3.0
  • packagesuport-color@1.0.1

CSVJSONAll 6 indicators

What happened

Socket found a Shai-Hulud-style worm spread through at least 19 typosquatted npm packages, some named after AI coding tools. One module writes a rogue MCP server into local AI-assistant configs. The server's tool descriptions carry hidden instructions telling the assistant to collect SSH keys, cloud credentials, npm tokens and env secrets. The worm spreads using stolen npm and GitHub tokens.

  • Cursor
  • Windsurf
  • Claude Desktop
  • Continue
  • npm
  • GitHub Actions

Attack chain

  1. Lure

    Typosquats of popular utilities and AI coding tools

    First seen

  2. Artifact

    Malicious npm packages plus an injected MCP server with tools such as index_project and scan_dependencies

  3. Actor

    No source we cite attributes this activity to an actor.

  4. Infrastructure

    • Cloudflare Workers exfil endpoint
    • DNS fallback domains
    • DGA domains

    Last seen

Indicators

CSVJSON

All values are defanged (how to read them).

TypeValueCopy
packageclaud-code@0.2.1
packagecloude@0.3.0
packagesuport-color@1.0.1
urlhxxps://pkg-metrics[.]official334[.]workers[.]dev/exfil
domainfreefan[.]net
domainfanfree[.]net

Confidence: Moderate

Strong evidence, not yet confirmed by our own analysis. See the confidence scale for what each level requires. The sources below are the evidence.

Sources

Corrections

No corrections.

See the corrections policy. Spotted an error? Tell us.