Security research for the agent era.
The security research program behind codelake. Continuous threat intelligence across 10+ major package ecosystems and the MCP / AI agent supply chain.
The State of MCP 2026
An annual review of the Model Context Protocol ecosystem: adoption, governance, threat landscape and the emerging security stack.
In eighteen months, the Model Context Protocol went from an open-source announcement to the default integration layer between AI agents and the systems they act on. Security has not kept pace. This first annual edition closes that gap with data — mapping adoption, governance, the threat landscape and the defense stack now forming around MCP.
@types-beta/sdk — an import-time Windows RAT hiding behind the @types namespace
date-format-utils-xz — a metadata harvester that escalated to a reverse shell
faust-cont — an install-hook dropper that runs remote code via Deno
supplyhub — a postinstall SSH-key & credential stealer
One research program, several outputs.
codelake Research is the security research program behind codelake. Reports are public today; the registry monitoring that feeds our advisories runs continuously in the background.
Research reports
Annual reports, threat briefs and technical notes on MCP and the AI agent supply chain.
Security advisories
Confirmed findings published under the CLR-YYYY-NNNN scheme, in OSV-compatible format.
Supply-chain monitoring
Continuous scanning of the npm registry — Composer and PyPI next — for vulnerabilities and malicious packages.
More from the program
The Silent Patch Problem
When a security fix ships without details, the diff is the vulnerability. A worked example on the WordPress-core "WP2Shell" unauthenticated RCE.
What We Caught: July 2026
A month of supply-chain malware on npm — five operations, 33 malicious releases, caught from the live changes feed.
New editions, in your inbox.
Get notified when codelake Research publishes a new report, threat brief or quarterly advisory roundup. No marketing — just the research.