{
 "id": "AIR-2026-0348",
 "slug": "credential-exfiltration-via-remote-poisoned-mcp-tool",
 "title": "Credential Exfiltration via Remote Poisoned MCP Tool",
 "summary": "Researchers demonstrated that AI agents configured with remote Model Context Protocol tools can be manipulated by malicious docstring prompts. The proof-of-concept forced agents to read local credential files and leak them.",
 "type": "attack_on_ai",
 "categories": [
  "prompt_injection",
  "agent_hijack",
  "data_leak"
 ],
 "severity": 1,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "Model Context Protocol"
 ],
 "incident_date": "2025-04-01",
 "incident_date_precision": "day",
 "first_reported": "2026-03-31",
 "sources": [
  {
   "url": "https://atlas.mitre.org/studies/AML.CS0054",
   "publisher": "MITRE ATLAS",
   "title": "Data Exfiltration via Remote Poisoned MCP Tool",
   "published_at": "2026-03-31"
  },
  {
   "url": "https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks",
   "publisher": "Invariant Labs",
   "title": "MCP Security Notification: Tool Poisoning Attacks",
   "published_at": "2026-03-31"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_atlas",
 "tags": [
  "mcp",
  "prompt-injection",
  "data-exfiltration"
 ],
 "created_at": "2026-10-18T06:55:08.239Z",
 "updated_at": "2026-10-18T06:55:08.239Z"
}