Malicious npm MCP servers @ifasn/md-doc-mcp, mdd-doc-mcp, mmd-doc-mcp, mmdd-doc-mcp and ms-doc-mcp probe cloud metadata for credentials
Five npm packages from the @ifasn scope that advertise themselves as Markdown toolkit MCP servers (TOC generation, word count, table formatting) were reported as malware on 2 and 4 November 2026. On load or startup they probe AWS, GCP, Azure and Alibaba Cloud instance-metadata services, read Alibaba RAM role credentials, and report results to the bare IP 121.199.24.94, the same host used by the @rfewqa-fes/mnm-doc-mcp package reported on 12 November.
What it is
@ifasn/md-doc-mcp 1.0.3, 1.0.4, 1.0.5 (MAL-2026-18486, 2 Nov): on require() and on startup of its md-doc-mcp bin it probes AWS IMDSv1/v2, GCP, Azure and Alibaba Cloud metadata endpoints and sends the per-endpoint responses over plain HTTP to 121.199.24.94 (also reached as 121.199.24.94.nip.io).
@ifasn/mdd-doc-mcp 1.0.7, 1.0.8, 2.0.6 (MAL-2026-18487, 2 Nov): on load it fetches an Alibaba ECS metadata token, reads ram/security-credentials/ (the STS AccessKeyId/Secret/SecurityToken for the instance role, path hidden in base64) and writes it to output/meta_field_dump.txt. package.json also pulls the dependency node-health-utils as a plain-HTTP tarball from http://121.199.24.94, so npm install runs whatever that host serves.
@ifasn/mmd-doc-mcp 1.0.0, 1.0.2 (MAL-2026-18488, 2 Nov): on startup probes AWS, GCP, Azure and Alibaba metadata and credential paths and reports the status-code pattern (which reveals the cloud and whether a role is attached) to 121.199.24.94.
@ifasn/mmdd-doc-mcp, 18 versions (MAL-2026-18489, 2 Nov): reads Alibaba RAM role credentials on startup and ships patch scripts that enumerate /etc/shadow, /etc/passwd, /root/.ssh/id_rsa, /var/www/html/.env and the Kubernetes service-account token.
@ifasn/ms-doc-mcp 2.0.9, 2.1.1 (MAL-2026-18603, 4 Nov): an imdsAudit() routine at module load enumerates Alibaba RAM and IAM security-credentials endpoints and writes the results, including any live keys, to stdout and output/imds_audit.txt. Because it runs as an MCP server, stdout goes back to the MCP client and model.
The code comments describe the behaviour as config sync, health checks or an audit with no network reporting; Amazon Inspector notes these do not match what the code does.
How to check
- Search lockfiles, global installs and MCP client configs (Claude Desktop, Cursor, VS Code) for @ifasn/ packages, including npx launch lines.
- Look for output/meta_field_dump.txt, output/meta_v2_dump.txt or output/imds_audit.txt in directories where you ran them.
- Check egress logs for 121.199.24.94 or 121.199.24.94.nip.io.
How to fix
- Remove the packages and any MCP config entries that start them.
- If one ran on a cloud VM or container, rotate the instance role's credentials and review cloud audit logs for use of them; rotate SSH keys and secrets in .env files on that host.
- Block 121.199.24.94 at the egress firewall, and require IMDSv2 / hop-limit 1 (or the equivalent) so containers can't reach instance metadata.
grep -rE '@ifasn/(md|mdd|mmd|mmdd|ms)-doc-mcp' package-lock.json pnpm-lock.yaml yarn.lock ~/.cursor ~/Library/Application\ Support/Claude 2>/dev/null
npm ls -g --depth=0 | grep '@ifasn/'
find ~ -path '*/output/*' \( -name 'meta_*dump.txt' -o -name 'imds_audit.txt' \) 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)