Endpoint · Shadow AI · Supply Chain

Endpoint Security for Developer
Devices and Shadow AI

Quick answer
TigerGate Endpoint Security is a lightweight agent for employee and developer devices. It inventories the AI tools, models and MCP servers people use, finds vulnerable or malicious packages, IDE extensions and browser extensions, and blocks risky installs before they run. It reports into the same TigerGate console as code, cloud and runtime findings. Complements your EDR; it does not replace it.
Start Free TrialShadow AI solution

Why Developer Devices Are the New Supply-Chain Front Door

Developer laptops hold the keys to the kingdom: GitHub and npm tokens, SSH keys, kubeconfigs, cloud credentials and session cookies. Three recent incidents show how attackers reach that keyring through packages and IDE extensions:

  • Shai-Hulud npm worm (Sep 2025) — self-propagating malware across hundreds of npm packages, exfiltrating environment variables and tokens via GitHub Actions.
  • GlassWorm on Open VSX (Oct 2025) — malicious VS Code extensions on the Open VSX registry stealing credentials and PATs.
  • Nx Console compromise (May 2026) — a popular JetBrains plugin trojaned via a maintainer account takeover.

EDR won't catch a maintainer-signed npm package installing a credential stealer in your node_modules. Endpoint Security governs what gets onto the laptop in the first place.

See Every AI Tool Your Teams Use

The endpoint agent inventories AI tools and surfaces the ones security teams need to approve or block:

AI desktop + browser apps

ChatGPT, Claude, Gemini, Perplexity and dozens more. Local and web.

IDE AI assistants

Cursor, Copilot, Claude Code, Continue, Cline. Per-IDE, per-user.

AI coding CLIs

Aider, Codex CLI, llm, cline-cli — invoked from terminal, often unlogged.

MCP servers (local + remote)

Inventories every MCP server configured in Claude Desktop, Cursor and VS Code, with permissions.

Local models

Ollama, LM Studio, llama.cpp — detects downloaded weights and running inferences.

Personal vs corporate accounts

Flags AI tools authenticated with personal accounts for data-governance review.

Block Malicious Packages and Extensions Before Install

TigerGate intercepts installs at the registry layer. Known-malicious packages, typosquats and compromised extensions are blocked before npm install, pip install or an extension marketplace install can run post-install scripts. Optional cooldowns delay fresh package versions by a configurable window so a worm can't propagate inside your fleet before anyone notices.

EcosystemPackage managersDetectBlock
npmnpm, yarn, pnpmYesYes
PyPIpip, poetry, uvYesYes
MavenMaven, GradleYesYes
NuGet.NET CLI, NuGetYesYes
Go modulesgo getYesPartial
RubyGemsgem, BundlerYesPartial
VS Code / Open VSX extensionsVS Code marketplaceYesYes
JetBrains pluginsJetBrains marketplaceYesPartial
Chrome / Edge extensionsChrome Web Store, Edge Add-onsYesYes

Vulnerable Dependencies on Laptops, Not Just in Repos

Repo-level SCA sees what's in your package.json. Endpoint SCA sees what's actually installed under ~/node_modules, ~/.local/pipx and global package stores — including dependencies pulled in by side projects, forks and local experiments that never land in a commit. Findings reuse the same vulnerability sources as TigerGate SCA/SBOM: OSV, GitHub Advisory, NVD.

Secrets Left on Devices

Scans local files, shell history, .env files and common config paths for credentials. Alerts route to the same workflow as repo-level secret scanning, with rotation guidance per secret type.

Policies, Approvals and MDM Rollout

  • Group policies by team, device OS, or role (e.g. "Platform engineers may install local models; Finance may not")
  • Request/approve flow for ad-hoc AI tools and risky packages
  • Deploy via Jamf, Microsoft Intune and Kandji with a signed installer package, or manually with a one-line installer
  • Silent install, no user interaction required

Privacy by Design: What Leaves the Device

DataSent to TigerGateStays on device
AI tool / MCP / package / extension inventoryYes (metadata only)—
Device identifier & OS versionYes—
Policy decisions (allow/block events)Yes—
Prompt or completion contents—Yes (never leaves)
File contents, source code—Yes
Browsing history—Yes

Connected to Code, Cloud and Runtime

A leaked GitHub token on a laptop ties to the repos it can read, the cloud role those repos can assume via GitHub Actions, and the production workloads that role touches. Endpoint findings route into the same security graph as code security, CSPM, CIEM and eBPF runtime, so one laptop finding can light up every downstream risk.

Compliance Evidence

  • SOC 2 CC6.8: malicious software prevention — package firewall + extension blocking
  • ISO 27001 A.8.7: protection against malware — install-time blocking across ecosystems
  • ISO 27001 A.8.19: installation of software on operational systems — approval workflow, device-level audit log
  • ISO 42001: AI inventory — AI tool, MCP server and model inventory for AI management system evidence

See compliance automation.

Endpoint Security FAQ

Developer endpoint security protects the laptops engineers use to write and ship code. It inventories AI tools, packages, IDE and browser extensions on each device, flags vulnerable or malicious components, and blocks risky installs before they can steal tokens, SSH keys or cloud credentials.
A lightweight TigerGate agent on each device identifies AI desktop and browser apps, IDE assistants, AI coding CLIs, MCP servers and the models they call, and shows which devices and teams use them. Security teams can then approve or block each tool.
Yes. Installs from supported package managers are checked against TigerGate threat intelligence, and known-malicious packages and extensions are blocked before their code runs. Supported ecosystems include npm, PyPI, Maven, NuGet and VS Code extensions.
No. It complements EDR. EDR looks for malicious behaviour on a device, while TigerGate governs what software, dependencies and AI tools get onto developer devices and connects those findings to your code, cloud and runtime risk.
Only inventory and finding metadata, such as tool names, package versions, device identifiers and policy decisions. Prompt contents, file contents and browsing history are not uploaded.
macOS, Windows and Linux. The agent can be rolled out through MDM tools such as Jamf, Microsoft Intune and Kandji, or installed manually.
Both protect developer devices from risky packages and shadow AI. TigerGate adds this to a platform that also covers cloud posture, Kubernetes and eBPF runtime protection, so device findings sit next to the cloud and workload risk they could lead to.
Per device per month, usage-based like every TigerGate module, with a 14-day free trial and no credit card.

Try Endpoint Security Free

Deploy the agent on 10 developer laptops in minutes. First shadow-AI and dependency findings inside the trial.

Start Free TrialCompare vs Aikido