🧰 Claude Desktop

The items that name it 25

• The OpenSourceMalware Show #23
  talkas affected products2026-10-02It is written here as Claude Desktop (macOS).

  Lookalike site claude-desktop.com served a signed macOS DMG installer (~210MB, about 4x the real ~50MB installer), hosted on GitHub, appearing to be the genuine app padded with an added malicious payload; Apple reportedly revoked the signing certificate after the report.

• graygnatconsole/mcp-audit-tool
  repoas affected products2026-09-26It is written here as Claude Desktop (claude_desktop_config.json).

  claude_desktop_config.json

• 30+ MCP CVEs in Two Months: What the Early-2026 Vulnerability Wave Means for Your MCP Supply Chain
  articleas affected products2026-09-23

  MCP client that used mcp-remote at the time of CVE-2025-6514

• The OpenSourceMalware Show #21
  talkas classification.technologies2026-09-17
• mukul975/cve-mcp-server
  repoas classification.technologies2026-09-07
• The AI Attack Surface: How Threat Actors Abuse Trusted AI Platforms
  writeupas classification.technologies2026-08-27
• IAM Policy Autopilot
  repoas tools2026-08-25

  AI assistant client that can be configured via claude_desktop_config.json to use the IAM Policy Autopilot MCP server

• [tl;dr sec] #342 - Figma's Agentic Detection, Agent Identity, Uber's Agent-(E)DR
  articleas affected products2026-08-20It is written here as Claude Desktop (Anthropic).

  Anthropic

• Orca Security: Shadow AI Governance
  articleas affected products2026-08-13It is written here as Claude Desktop (listed in Microsoft's shadow AI preview).

  listed in Microsoft's shadow AI preview

• GhostJacking Attacks: AI Agent Takeover via Firewall Evasion
  writeupas affected products2026-08-10

  Egress sandbox / Envoy proxy JWT-based allowed_hosts allowlist; zero-day JWT cross-reuse flaw (container_id not validated) allows sandbox escape and egress-control bypass; reported to Anthropic, confirmed and patched before publication, no CVE assigned

• RamanMG/Burp-MCP-Unrestricted
  repoas affected products2026-07-29

  configured as an MCP client via installer-written config

• Inside FakeAgent: How a Claude Desktop Malvertising Campaign Hit 29 Organizations with SectopRAT
  writeupas classification.technologies2026-07-22
• Security notice and deprecation: standalone circleci-mcp-server
  advisoryas tools2026-07-20

  AI tool that connects to the CircleCI MCP server via claude_desktop_config.json

• AI Security Briefing
  articleas classification.technologies2026-07-18
• AI Security Briefing
  writeupas affected products2026-07-18

  registers a claude:// URL scheme that auto-submits a prompt from a clicked link without user confirmation or review; fixed in version 1.1.2321

• jscrambler npm Package Compromised in Supply Chain Attack
  writeupas affected products2026-07-11

  Config files .config/Claude/claude_desktop_config.json, .claude.json targeted for API keys/MCP credentials

• Advanced Prompt Injection Techniques 2026: 7 Attack Chains
  articleas affected products2026-07-04

  Uses persistent long-term memory that can be exploited for multi-turn injection persistence; runs MCP tools

• Teycir/SkillsGuard
  repoas affected products2026-06-21

  Platform SkillsGuard's MCP server and skill/SKILL.md integrate with, on macOS and Windows.

• i8void/cybersleuth
  repoas classification.technologies2026-05-14
• h1-brain
  repoas tools2026-05-06

  MCP-compatible client used to interact with h1-brain

• mcpwn
  repoas affected products2026-05-06

  mcpwn can scan all MCP servers from Claude Desktop config via --claude-config

• PatrikFehrenbach/h1-brain
  repoas tools2026-05-05

  MCP-compatible client used to connect to h1-brain server

• The Vulnerable MCP Project — MCP vulnerability database
  repoas tools

  AI assistant client used to demonstrate GitHub MCP data leak

• Claude Desktop (macOS): malicious file in a Cowork folder could run commands on the host
  advisoryas affected productsIt is written here as Claude Desktop (macOS).

  File type omitted from block list allowed a malicious file in a Cowork folder to run commands on the host when opened

• Amazon Kiro Prompt Injection Can Exfiltrate Sensitive Data Through Kiro Powers
  articleas affected products

  PromptFiction vulnerability, single-click link tricks agent into harvesting data or running arbitrary code

In every workspace you can read 39