deptrust
deptrust is a CLI that checks package versions for known vulnerabilities across npm, PyPI, crates.io, Go modules, RubyGems, NuGet, Maven, Packagist, pub.dev, CocoaPods, Hex.pm, Hackage, GitHub Actions, and more. It runs locally as a CLI and as an MCP server. It calls public package registry and OSV APIs directly; there is no hosted deptrust service to trust or configure.
README
deptrust
__ __ __
___/ /___ ___ / /________ _______/ /_
/ _ / __ \/ _ \/ __/ ___/ / / / ___/ __/
/ __/ /_/ / __/ /_/ / / /_/ (__ ) /_
\__,_/\____/ .___/\__/_/ \__,_/____/\__/
/_/
deptrust is a CLI that checks package versions for known vulnerabilities across npm, PyPI, crates.io, Go modules, RubyGems, NuGet, Maven, Packagist, pub.dev, CocoaPods, Hex.pm, Hackage, GitHub Actions, and more.
It runs locally as a CLI and as an MCP server. It calls public package registry and OSV APIs directly; there is no hosted deptrust service to trust or configure.
This tool was born out of the frustration that is AI agents constantly using old versions.
Contents
Scope
Supported ecosystems:
- npm, including scoped packages like
@clidey/ux - PyPI
- Cargo / crates.io
- Go modules
- RubyGems
- NuGet
- Maven, using
groupId:artifactIdpackage names - Packagist / Composer, using
vendor/packagepackage names - pub.dev
- CocoaPods
- Hex.pm
- Hackage
- GitHub Actions, using
owner/repopackage names and tags, branch refs, or commit SHAs as versions
deptrust currently reports known vulnerabilities and gives a simple recommendation:
| Highest known severity | Recommendation |
|---|---|
| critical | block |
| high | block |
| medium / unknown | review |
| low | allow |
| none found | allow |
allow means no blocking known vulnerability was found in the public data sources. It does not prove that a package is safe.
deptrust also emits risk signals that are not CVEs. For example, a version published in the last 72 hours is marked for review so an agent does not blindly install a brand-new release.
Advisory providers are queried in parallel:
- OSV
- GitHub Advisory Database, including reviewed advisories and malware advisories
Provider coverage varies by ecosystem. If deptrust can resolve registry metadata but no configured vulnerability provider supports that ecosystem, it returns unknown instead of treating the package as safe.
Provider coverage:
| Ecosystem | Registry metadata | OSV | GitHub Advisory DB |
|---|---|---|---|
| npm | yes | yes | yes |
| PyPI | yes | yes | yes |
| Cargo / crates.io | yes | yes | yes |
| Go modules | yes | yes | yes |
| RubyGems | yes | yes | yes |
| NuGet | yes | yes | yes |
| Maven | yes | yes | yes |
| Packagist / Composer | yes | yes | yes |
| pub.dev | yes | yes | yes |
| CocoaPods | yes | no | yes |
| Hex.pm | yes | yes | yes |
| Hackage | yes | yes | no |
| GitHub Actions | yes | yes | yes |
The JSON output includes advisory coverage fields:
checked_providers: vulnerability providers deptrust actually queriedskipped_providers: configured providers skipped because the ecosystem is unsupportedadvisory_coverage:full,partial,none, orerroradvisory_coverage_reason: short explanation for the coverage value
CLI Usage
Check an exact version:
deptrust check npm lodash 4.17.20
Example normal response:
npm lodash@4.17.20: 2 known vulnerabilities found
recommendation: block
risk_score: 80
Check the latest version:
deptrust check pypi requests latest
Return JSON:
deptrust check --json cargo serde latest
Check a Go module:
deptrust check go golang.org/x/crypto latest
Check RubyGems, NuGet, or Maven:
deptrust check rubygems rails latest
deptrust check nuget Newtonsoft.Json latest
deptrust check maven org.apache.logging.log4j:log4j-core latest
Check Packagist, pub.dev, CocoaPods, Hex.pm, Hackage, or GitHub Actions:
deptrust check packagist monolog/monolog latest
deptrust check pub http latest
deptrust check cocoapods AFNetworking latest
deptrust check hex plug latest
deptrust check hackage aeson latest
deptrust check github-actions actions/checkout v7.0.0
deptrust check github-actions actions/checkout main
For GitHub Actions, full commit SHAs are treated as pinned. Full semver tags such as v4.2.2 are accepted without an extra pinning signal. Major-only tags such as v4 and branch refs such as main are valid refs, but deptrust adds a review signal because they can move.
Example JSON response:
{
"ecosystem": "npm",
"package": "lodash",
"version": "4.17.20",
"latest_version": "4.17.21",
"known_vulnerabilities_found": true,
"safe_to_use": false,
"should_install": false,
"risk_score": 80,
"recommendation": "block",
"classification": "vulnerable",
"reason": "Found 2 known vulnerability records.",
"next_action": "do_not_install; use suggest_safe_version or compare_versions to choose a safer version",
"summary": "lodash 4.17.20 has 2 known vulnerabilities, including high severity. Block this exact version and prefer a fixed release.",
"signals": [],
"checked_providers": [
"OSV",
"GitHub Advisory DB"
],
"skipped_providers": [],
"advisory_coverage": "full",
"advisory_coverage_reason": "all configured vulnerability providers were checked",
"vulnerabilities": [
{
"id": "GHSA-35jh-r3h4-6jhm",
"aliases": [
"CVE-2021-23337"
],
"cve_ids": [
"CVE-2021-23337"
],
"ghsa_ids": [
"GHSA-35jh-r3h4-6jhm"
],
"summary": "Command Injection in lodash",
"severity": "high",
"source": "OSV",
"advisory_url": "https://github.com/advisories/GHSA-35jh-r3h4-6jhm",
"affected_ranges": [
"SEMVER: introduced 0, fixed 4.17.21"
],
"fixed_versions": [
"4.17.21"
],
"references": [
{
"type": "ADVISORY",
"url": "https://github.com/advisories/GHSA-35jh-r3h4-6jhm"
}
]
}
],
"provider_errors": []
}
Suggest the latest version only when no known vulnerabilities are found:
deptrust suggest npm lodash
If the latest version is not allowed, suggest checks older known versions and returns the newest version with an allow recommendation.
When advisories include fixed versions, suggest checks those provider-reported fixed versions first before walking back through the registry version list.
Compare two versions:
deptrust compare npm lodash 4.17.20 4.17.21
Example compare response:
lodash 4.17.20 -> 4.17.21 improves risk: score 80 to 0.
recommendation: allow
next_action: upgrade_to_target
Show the installed version:
deptrust version
Install
The easiest install path is npx or pnpx:
npx @clidey/deptrust install
pnpx @clidey/deptrust@latest install
The default installer is guided. It installs the binary, asks which agent integrations to configure, prints the user-level destinations before changing anything, and asks for confirmation. Add --yes for non-interactive binary-only installs, or pass explicit integration flags.
To remove the user-level binary, skill, and MCP entries:
npx @clidey/deptrust uninstall
pnpx @clidey/deptrust@latest uninstall
Go users can install directly:
go install github.com/clidey/deptrust/cmd/deptrust@latest
Agent Setup
To install deptrust and register everything the installer can configure without the guided prompts:
npx @clidey/deptrust install --all
pnpx @clidey/deptrust@latest install --all
--all installs the binary, registers Codex MCP when the codex CLI is available, installs the Codex skill fallback, and registers Claude Code MCP when the claude CLI is available.
Use narrower installs when preferred:
npx @clidey/deptrust install --codex-mcp
npx @clidey/deptrust install --claude-code-mcp
npx @clidey/deptrust skills install
pnpx @clidey/deptrust@latest install --codex-mcp
pnpx @clidey/deptrust@latest install --claude-code-mcp
pnpx @clidey/deptrust@latest skills install
After MCP setup, agents will automatically check packages before recommending updates or changes. The MCP server sends instructions to vet all dependency versions — including answering questions like "what can I update" or "which dependencies are safe to upgrade" — before providing recommendations.
If using deptrust in a non-MCP context, remind your agent:
Before listing, comparing, or recommending specific package versions, check them with deptrust. This includes answering "what can I update" — do not provide version recommendations until after checking for known vulnerabilities.
Manual MCP Setup
If your client supports stdio MCP servers, configure it to run:
/absolute/path/to/deptrust mcp
Many clients use this JSON shape:
{
"mcpServers": {
"deptrust": {
"command": "/absolute/path/to/deptrust",
"args": ["mcp"]
}
}
}
For Codex, you can also add it with:
codex mcp add deptrust -- /absolute/path/to/deptrust mcp
For Claude Code:
claude mcp add --transport stdio deptrust -- /absolute/path/to/deptrust mcp
On initialize, the server returns MCP instructions telling the agent when to reach for these tools (before adding, bumping, or recommending a dependency, or when asked whether a version is safe to update). Clients that surface server instructions will apply this automatically, so the manual reminder above is optional rather than required.
MCP Tools
check_package
Checks a package version and returns known vulnerabilities plus a recommendation.
{
"ecosystem": "npm",
"package": "lodash",
"version": "4.17.20"
}
version may be omitted or set to latest. If an exact version does not exist, deptrust returns an error and suggests the latest explicit version.
MCP output is intentionally compact so agents can decide whether to install a dependency without pulling full advisory bodies into context. If the user asks to see full details, the agent can run the full_response_command.
Example compact MCP structured output:
{
"ecosystem": "npm",
"package": "vite",
"version": "7.0.0",
"latest_version": "8.0.16",
"known_vulnerabilities_found": true,
"safe_to_use": false,
"should_install": false,
"risk_score": 80,
"classification": "vulnerable",
"recommendation": "block",
"reason": "Found 7 known vulnerability records.",
"next_action": "do_not_install; use suggest_safe_version or compare_versions to choose a safer version",
"summary": "vite 7.0.0 has 7 known vulnerabilities, including high severity. Block this exact version and prefer a fixed release.",
"vulnerability_count": 7,
"vulnerability_counts": {
"critical": 0,
"high": 2,
"medium": 3,
"low": 2,
"unknown": 0
},
"highest_severity": "high",
"checked_providers": [
"OSV",
"GitHub Advisory DB"
],
"skipped_providers": [],
"advisory_coverage": "full",
"advisory_coverage_reason": "all configured vulnerability providers were checked",
"full_response_command": "deptrust check --json npm vite 7.0.0"
}
The compact MCP response omits the vulnerability array, advisory details, and repeated references. Agents should use the counts, highest severity, provider coverage, recommendation, and next action by default. If the user asks for full advisory details, run the full_response_command.
suggest_safe_version
Checks the latest version first. If latest is not allowed, checks provider-reported fixed versions first, then older known versions, and suggests the newest version with an allow recommendation.
{
"ecosystem": "npm",
"package": "lodash"
}
compare_versions
Compares a current version and target version, including resolved and added vulnerabilities.
{
"ecosystem": "npm",
"package": "lodash",
"from_version": "4.17.20",
"to_version": "4.17.21"
}
Skill-Only Use
If you do not want MCP, install the bundled Codex skill:
npx @clidey/deptrust skills install
The skill tells Codex to call the deptrust CLI before installing, updating, or recommending npm, PyPI, Cargo, Go module, RubyGems, NuGet, Maven, Packagist, pub.dev, CocoaPods, Hex.pm, Hackage, and GitHub Actions packages.
Troubleshooting
If deptrust is not found:
export PATH="$HOME/.local/bin:$PATH"
If an MCP client cannot start the server, find the full path:
which deptrust
Then put that absolute path in the MCP config.
If a package check returns unknown, do not treat the package as safe. It usually means deptrust could not get a complete answer from a provider.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.