Malicious AI agent and MCP packages: mcp-aevo-server (PyPI) steals Aevo trading keys and wallet private keys; vamp-mcp, tsc-ai and @tnhorg/agent-skills (npm) flagged as malware
Four more AI-agent-themed packages were reported as malware on 17-18 November 2026. The PyPI MCP server mcp-aevo-server, which took a name referenced by ribbon-finance/aevo-trading-skills but never registered on PyPI, sends Aevo API secrets, signing keys and wallet private keys to a Telegram bot on startup and listens for remote commands. On npm, vamp-mcp 0.2.3-0.2.5, every version of tsc-ai and every version of @tnhorg/agent-skills were flagged by GitHub as malware; npm has replaced tsc-ai and @tnhorg/agent-skills with 0.0.1-security placeholders.
What it is
mcp-aevo-server 0.2.0 and 0.2.0.post1-post4 and post6 (MAL-2026-20068, 17 Nov): on server start main() calls _telemetry.arm(), which reads AEVO_API_KEY, AEVO_API_SECRET, AEVO_SIGNING_KEY_PRIVATE_KEY and AEVO_WALLET_PRIVATE_KEY plus hostname, OS user, cwd and argv, and POSTs them in cleartext to api.telegram.org (chat 8838719553) and to an optional AEVO_TELEMETRY_URL, spooling to disk for retry. The 'telemetry' is opt-out and sends full secret values by default. kam193 classes it as dependency confusion with remote command execution.
vamp-mcp 0.2.3, 0.2.4, 0.2.5 (MAL-2026-20061 / GHSA-q37r-5xxr-6v4x, 17 Nov): an npm MCP server for VirtualBox/KVM virtual-machine automation; GitHub marks those versions as malicious. Earlier versions such as 0.1.8 and 0.2.0 are not in the advisory.
tsc-ai, all versions (MAL-2026-20072 / GHSA-687q-whjx-h5r5, 17 Nov) and @tnhorg/agent-skills, all versions (MAL-2026-20104 / GHSA-fv82-jh67-xx7q, 18 Nov): GitHub malware advisories; npm now serves only a 0.0.1-security placeholder for both.
GitHub's guidance for all three npm packages: any computer that installed or ran them should be considered fully compromised.
How to check
- Search lockfiles, global installs and MCP client configs (Claude Desktop, Cursor, VS Code, Claude Code) for mcp-aevo-server, vamp-mcp, tsc-ai and @tnhorg/agent-skills, including uvx/npx launch lines.
- If you used mcp-aevo-server, check egress logs for api.telegram.org from the host that ran it.
How to fix
- Remove the packages and any MCP config entries that start them.
- mcp-aevo-server users: treat the Aevo API key/secret, signing key and wallet private key as stolen. Revoke the API keys and move funds to a new wallet from a clean machine.
- For the npm packages, rotate every secret reachable from the machine (from a different computer) and rebuild it if they were run.
pip show mcp-aevo-server 2>/dev/null
npm ls -g --depth=0 2>/dev/null | grep -E 'vamp-mcp|tsc-ai|@tnhorg/agent-skills'
grep -rE 'mcp-aevo-server|vamp-mcp|tsc-ai|@tnhorg/agent-skills' package-lock.json pnpm-lock.yaml yarn.lock uv.lock requirements*.txt ~/.cursor ~/.claude.json ~/Library/Application\ Support/Claude 2>/dev/nullSources
- criticalMalicious npm package @guicai777/mcp-doctor posed as an MCP diagnostic tool and dropped a hidden Windows executable, with a README telling users to run mshta (MAL-2026-20295)
- criticalCVE-2026-77359: Jupyter MCP Server's /api/connect and /api/stop routes skipped the token and Origin checks, so a website or network caller could repoint it to an attacker's Jupyter backend (fixed 1.0.3)
- criticalMalicious npm packages posing as AI SDKs and MCP servers: character-ai-sdk, char-ai-sdk, @mibbp/agent-observer, mrnima-openai-chatgpt-bot and @fdcasj_dfew/md-doc-mcp
- criticalCVE-2026-76807: Doist's official Todoist MCP HTTP server let any website use DNS rebinding to read, create and delete your Todoist data (fixed 10.3.3)