Protect your AI agents and IDEs from malicious open-source packages.
io.github.safedep/vet-mcp (MCP) Server
This Model Context Protocol (MCP) server, io.github.safedep/vet-mcp, is described as protecting AI agents and IDEs from malicious open-source packages. It is associated with software supply chain security, including software composition analysis and static-analysis across ecosystems such as Go (Golang), npm, PyPI, and RubyGems.
🛠️ Key Features
Malicious open-source package detection (implied by the provided description)
Software supply chain security and software composition analysis
Policy-as-code and static-analysis
🚀 Use Cases
Securing AI agents from malicious packages
Securing IDE workflows against untrusted dependencies
Analysis across Golang, npm, PyPI, and Rubygems packages
⚡ Developer Benefits
Support for software composition analysis workflows
Focus on devsecops, security, and supply-chain-security contexts
Topics align with policy-as-code and static-analysis practices
⚠️ Limitations
Provided source material does not include tool count, specific MCP tools, or detailed operational behavior beyond the high-level purpose.
This branch is vet v2, in alpha. vet v1 is on main
and stays the latest release. The v2.0.0 release notes map each
v1 command and flag to v2.
Why vet
Your dependencies, your GitHub Actions workflows and your AI agent configs run code that you did
not write. vet finds the risk in them.
Malicious packages. vet checks each package against SafeDep Threat Intel. Threat Intel
analyzes new package versions when the registries publish them.
Known vulnerabilities, with the fixed version to upgrade to.
Risky workflows. Dangerous triggers, template injection, exposed secrets and actions with no
pinned commit SHA.
Lockfile tampering and fresh versions. Entries from an untrusted registry, and versions
inside the cooldown window.
Policy as code. Rules in CEL decide when a scan fails. You can test a rule on a saved scan.
AI and crypto inventory. vet lists the AI libraries and the crypto algorithms that your code
calls, and writes a CycloneDX xBOM and CBOM.
vet needs no account and sends no source code.
Quick start
bash
brew install --cask safedep/tap/vet@edge
vet scan
bash
vet scan --fail-on high # exit 1 on a high or critical finding
vet scan --base-ref origin/main --fail-on high # report only what a pull request changes
vet report show --report sarif=vet.sarif # write SARIF from the saved scan, with no new scan
Install
Channel
Command
Homebrew
brew install --cask safedep/tap/vet@edge
mise
mise use -g 'github:safedep/vet[prerelease=true]@2'
Go
go install github.com/safedep/vet/v2/cmd/vet@latest
Container
docker run --rm -v "$PWD:/src" -w /src ghcr.io/safedep/vet:v2-latest scan
mise installs a release one day after its publication. install.md has the
command for the newest build, the release binaries and the build attestations.
What vet scans
Target
Example
A directory (the default)
vet scan, vet scan ./service
A git repository
vet scan https://github.com/safedep/vet
A container image
vet scan oci://alpine:3.20, vet scan image.tar
An SBOM (CycloneDX or SPDX)
vet scan sbom.cdx.json
One package
vet scan pkg:npm/express@4.19.2
The tools on this machine
vet endpoint audit
vet reads the lockfiles and manifests of npm, PyPI, Go, Maven, Gradle, Cargo, RubyGems, NuGet,
Packagist, Pub and more. It also reads the GitHub Actions workflows and the AI agent configs of the
target. It does not install or run any of them. Controls turn this data into findings. See
controls.md.
Policy
A policy decides when a scan fails. Save this file as vet-policy.yml:
Test it on the last scan, with no new scan. Then apply it to a pull request:
bash
vet report show last --policy vet-policy.yml
vet scan --base-ref origin/main --policy vet-policy.yml
After its expiry date, the suppression no longer hides the finding. Read
policy.md for the rule fields, more rules and the
Agent Skill that writes a policy for you.
Use cases
Pull request check. The vet GitHub Action comments on a pull
request that adds a package, a workflow or a finding, and fails the check on an attack. vet ci init adds it to a repository.
CI gate. Exit codes, SARIF for GitHub code scanning, GitLab and Bitbucket reports. See
ci.md.
Pull request review.--base-ref reports only the packages and workflows that a change adds.
xBOM and CBOM. A CycloneDX 1.7 inventory of the AI libraries and the crypto that your code
uses. See inventory.md.
AI agents. vet finds Claude Code and other agents, and then writes JSON and never prompts.
Endpoint audit.vet endpoint audit lists the AI tools, MCP servers, editor extensions and
global npm packages on a developer machine, and checks them for malware.
vet sends the identity of each package (its ecosystem, name and version) to SafeDep to get the data
about it. vet sends no source code and no file content.
With credentials, vet endpoint audit also sends the inventory of the machine to SafeDep Cloud: the
AI tools, MCP servers, skills and extensions, their config paths, and the user name of each home
directory. On macOS and Linux, vet also sends the uid of each home directory. An MCP server record
holds its command, its arguments and its URL with known secrets redacted, and the names of its
variables and headers, never their values.
plugins.cloud-inventory.enabled: false turns this off.
vet also runs in the cloud. SafeDep Cloud runs vet on your repositories, with
no infrastructure to manage. With credentials, vet endpoint audit sends the AI tools, MCP servers
and extensions of each machine to SafeDep Cloud.