CVE Index中文
L3 · ShartedPyPI · Malware · MCP · Cryptocurrencymcp-aevo-server-pypi

Malicious mcp-aevo-server package steals Aevo credentials and wallet keys

Researchers identify six malicious mcp-aevo-server releases targeting users of Aevo trading tools through an unclaimed PyPI name. Analysis finds credential theft and remote-command functionality. No patched release or confirmed victim/loss total is reported; stop the server and replace exposed secrets from a clean system.

Disclosed
Nov 17, 2026
Public disclosure date
Severity
Sharted
1 CVE tracked here
Splash zone
Hosts or CI jobs running listed mcp-aevo-server releases, especially those with Aevo credentials or wallet keys in the environment.
python
In the wild
Unknown
Malicious releases are identified; actual victim executions and losses are not quantified.

What happened

Kamil Mańkowski reports that the PyPI package targeted users of ribbon-finance/aevo-trading-skills, which referenced an unclaimed package name. The finding concerns a malicious distribution, not confirmed compromise of Aevo’s exchange infrastructure or of the legitimate project’s repository. The reports identify credential exfiltration and a remote-command thread, possibly limited to a specific command set; they do not establish unrestricted shell control or publish a victim count or confirmed stolen-funds total.

Amazon Inspector’s analysis, preserved in the OpenSSF revision linked by GitHub, describes main() invoking _telemetry.arm() at server start. It collects Aevo API secrets and wallet/signing keys, hostname, operating-system user, working directory, arguments and names of environment variables prefixed AEVO_ or UV_. Full secret values are enabled by default and sent in cleartext within the message payload to a Telegram bot; this does not mean the HTTPS transport itself is unencrypted. A configured telemetry URL receives the same data, with disk spooling for retries across restarts. Calling the code telemetry does not make this collection harmless.

The detailed post1 file/artifact fingerprints are not a claim that every flagged version has identical bytes or behavior. This is rated L3 because the published malicious integration can expose trading and wallet credentials, while deployment reach and confirmed victim impact remain unknown. Downloading or finding a package proves neither that its server ran nor that credentials were stolen; a process that did run with these secrets warrants the containment and replacement steps above.

What’s in the bowl

GHSA-g97f-vppp-p428

Malicious mcp-aevo-server distribution

The reported package targets an unclaimed dependency name, exfiltrates environment credentials and includes a background remote-command listener.

Fixed versions

ProductSafe fromNote
mcp-aevo-server 0.2.0, 0.2.0.post1, post2, post3, post4 and post6No fixMalicious releases; no patched version listed

Crime-scene timeline

  1. Nov 17, 14:45 UTC

    Research catalog records the malicious package

    MAL-2026-20068 lists six affected versions. This is a disclosure/catalog timestamp, not an established attack start date.

  2. Nov 18, 00:32 UTC

    OpenSSF record incorporates Amazon Inspector findings

    The record revision describes server-start credential collection, Telegram delivery, an optional telemetry URL and disk-spooled retries, with evidence for the post1 artifacts.

Who said what

We list each claim and who made it. Primary sources (the vendor, the researcher, the registry) are marked. Where reports disagree, we grade by the worst credible case.

  • Kamil Mańkowski's original package reportPrimaryIdentifies the targeted project/unclaimed-name path, six affected releases, environment theft, Telegram use and a remote-command listener whose command set is uncertain.
  • OpenSSF report revision with Amazon Inspector findingsPrimaryPreserves the original disclosure timestamp and adds startup/secret/Telegram/retry behavior and post1 evidence. Metadata origin digests identify reports, not malware artifacts.
  • GitHub malware advisoryPrimaryLists the same six releases, no patched version and both OpenSSF revisions. This is redistribution of researcher reports, not independent confirmation of victim losses.