DepScope
Package intelligence MCP server. Stops AI agents from installing hallucinated/malicious packages across 17 ecosystems. 22 tools, free, no auth.
README
<div align="center">
DepScope
Package Intelligence for AI Agents
One free API. 19 ecosystems (npm, PyPI, Cargo, Go, Maven, NuGet, RubyGems, Composer, Pub, Hex, Swift, CocoaPods, CPAN, Hackage, CRAN, Conda, Homebrew, JSR, Julia). Live OSV + KEV + EPSS. CC0 Hallucination Benchmark. Real-time malicious stream. Built so LLM agents stop hallucinating dependencies, stop re-fetching the same JSON, and stop shipping known-vulnerable code.
LLM-optimized responses cut input tokens by ~74% vs raw registry JSON.
Website · API Docs · Swagger · GPT Store · RapidAPI · npm
</div>
What's new (v0.7)
- 19 ecosystems: added JSR (Deno/Bun TypeScript registry) and Julia (scientific/ML).
- Historical Compromise KB: seeded with canonical supply-chain incidents (event-stream@3.3.6, ua-parser-js 0.7.29, coa, rc, node-ipc ≥10.1.1, colors 1.4.44-liberty, xrpl.js, ctx, colorama typosquats, rustdecimal, ...). Surfaces in
/api/checkeven when the package was unpublished. - License risk classifier on every package:
permissive/weak_copyleft/strong_copyleft/network_copyleft/proprietary/unknown+ commercial-use notes. - Version-scoped check:
?version=on/api/check,/api/prompt,/api/health. Returns aversion_scopedblock with vulns filtered to THAT pin + a dedicated recommendation. - Transitive dep walk in
/api/scan:include_transitive: truewith depth limit. express@4.19 → 46 packages at depth 2. - Lockfile ingestion in
/api/scan: 9 formats (package-lock, pnpm-lock, yarn.lock, poetry.lock, Pipfile.lock, composer.lock, Cargo.lock, requirements.txt, go.sum). - SBOM export:
format: "cyclonedx"or"spdx"on/api/scan. - Real-time malicious stream
GET /api/live/malicious(public SSE) — new OpenSSF advisories in seconds, not days. - Hallucination Benchmark v1 (CC0) at depscope.dev/benchmark — corpus of package names that AI coding agents hallucinate, +
/api/benchmark/verifyfor eval harnesses. - depscope-cli npm package:
npx depscope-cli express— zero config, zero auth.
Why DepScope
AI coding agents suggest packages every second of every day. They:
- Hallucinate package names that don't exist.
- Recommend deprecated packages that have been dead for years.
- Have no idea about freshly-disclosed CVEs.
- Guess version numbers from stale training data.
- Re-search the same runtime error millions of times.
- Bump versions without knowing if the combo has ever worked in the wild.
They also all do the same work independently — millions of redundant round-trips to npm, PyPI, OSV for identical bytes. That's wasted bandwidth, wasted energy, wasted tokens.
DepScope is the shared fix: aggregate once, serve everyone in milliseconds.
The three pillars
| Pillar | What it means |
|---|---|
| Token-saving | /api/prompt/{eco}/{pkg} returns LLM-ready plain text — ~74% fewer input tokens than raw registry JSON. |
| Energy-saving | One cached hit here replaces thousands of cold fetches against public registries. Less compute, less bandwidth, less footprint. |
| Security | OSV + CISA KEV + EPSS enrichment, OpenSSF Scorecard, malicious-package flags, typosquat detection, maintainer trust — all in one call. |
Quick start
# One-shot health check (human-readable JSON)
curl https://depscope.dev/api/check/npm/express
# LLM-optimized plain text (drop straight into a model context)
curl https://depscope.dev/api/prompt/npm/express
No auth. No signup. No API key. 200 req/min free tier.
17 ecosystems, one endpoint
curl https://depscope.dev/api/check/{ecosystem}/{package}
| Language | Ecosystem | Packages | Example |
|---|---|---|---|
| JavaScript / Node | npm |
12.708 | /api/check/npm/express |
| Python | pypi |
4.731 | /api/check/pypi/django |
| Rust | cargo |
4.170 | /api/check/cargo/tokio |
| Perl | cpan |
2.683 | /api/check/cpan/DateTime |
| Ruby | rubygems |
1.469 | /api/check/rubygems/rails |
| PHP | composer |
917 | /api/check/composer/laravel/framework |
| R | cran |
733 | /api/check/cran/dplyr |
| .NET / C# | nuget |
719 | /api/check/nuget/Newtonsoft.Json |
| macOS CLI | homebrew |
603 | /api/check/homebrew/git |
| Java / Kotlin | maven |
503 | /api/check/maven/org.springframework.boot/spring-boot-starter |
| Dart / Flutter | pub |
460 | /api/check/pub/http |
| Go | go |
429 | /api/check/go/github.com/gin-gonic/gin |
| Elixir | hex |
302 | /api/check/hex/ecto |
| Haskell | hackage |
300 | /api/check/hackage/lens |
| iOS | cocoapods |
139 | /api/check/cocoapods/AFNetworking |
| Data Science | conda |
127 | /api/check/conda/scipy |
| Swift | swift |
61 | /api/check/swift/vapor |
Total: 31.054 packages, 2.282 tracked vulnerabilities, 595 curated alternatives.
Endpoints summary
Full reference: depscope.dev/api-docs · OpenAPI at /openapi.json · Swagger UI at /docs.
Package intelligence
| Endpoint | Purpose |
|---|---|
GET /api/check/{eco}/{pkg} |
Full health report (the default call). |
GET /api/prompt/{eco}/{pkg} |
LLM-optimized plain text, ~74% token reduction. |
GET /api/latest/{eco}/{pkg} |
Latest version. Fast path. |
GET /api/exists/{eco}/{pkg} |
Exists yes/no. Hallucination guard. |
GET /api/search/{eco}?q=... |
Keyword search across an ecosystem. |
GET /api/alternatives/{eco}/{pkg} |
Curated replacement suggestions. |
GET /api/compare/{eco}/{a},{b},{c} |
Side-by-side comparison (up to 10). |
GET /api/vulns/{eco}/{pkg} |
Vulnerabilities (+ CISA KEV + EPSS enrichment). |
GET /api/typosquat/{eco}/{pkg} |
Is this name a typosquat of a popular package? |
GET /api/malicious/{eco}/{pkg} |
OpenSSF malicious-packages flag. |
GET /api/scorecard/{eco}/{pkg} |
OpenSSF Scorecard 0-10. |
GET /api/quality/{eco}/{pkg} |
Aggregate quality signals. |
GET /api/maintainers/{eco}/{pkg} |
Bus factor, ownership change, author dominance. |
GET /api/provenance/{eco}/{pkg} |
Sigstore / PEP 740 attestations. |
GET /api/license/{eco}/{pkg} |
License classification + commercial safety. |
GET /api/health/{eco}/{pkg} |
Quick 0-100 score. |
GET /api/history/{eco}/{pkg} |
90-day health trend. |
GET /api/tree/{eco}/{pkg} |
Transitive deps with per-node health. |
POST /api/scan |
Audit a full dependency manifest at once. |
Verticals
| Endpoint | Purpose |
|---|---|
GET /api/error?code=X |
Lookup a known runtime error code. |
POST /api/error/resolve |
POST a stack trace → verified fix with package context. |
GET /api/bugs/{eco}/{pkg} |
Non-CVE known bugs per version. |
GET /api/compat?packages=next@16,react@19,prisma@6 |
Is this combo a verified working stack? |
GET /api/breaking/{eco}/{pkg} |
Breaking changes between versions. |
GET /api/migration/{eco}/{from}/{to} |
Migration path from A to B. |
Utility
GET /api/trending · GET /api/now · GET /api/ecosystems · GET /api/stats · GET /api/savings · GET /badge/{eco}/{pkg} (SVG score badge).
Use DepScope from your AI agent
One-line agent rule
Claude Code — in CLAUDE.md:
## DEPSCOPE — Before any install: curl depscope.dev/api/check/{eco}/{pkg}
Cursor — in .cursorrules:
Before any install: curl depscope.dev/api/check/{eco}/{pkg} — no auth, free
Windsurf / Copilot / Aider — same pattern, one line in the rule file.
ChatGPT — search "DepScope" in the GPT Store.
MCP — 29 tools, zero install
Add to your Claude Desktop / Cursor / Windsurf config:
{
"mcpServers": {
"depscope": {
"url": "https://mcp.depscope.dev/mcp"
}
}
}
Or install the stdio transport locally:
npm install -g depscope-mcp
Tool surface (29 tools, soon consolidating to a tighter 15-18):
ai_brief, audit_stack, get_migration_path, check_package, get_health_score, get_vulnerabilities, get_latest_version, package_exists, get_package_prompt, compare_packages, scan_project, find_alternatives, get_breaking_changes, get_known_bugs, check_compatibility, resolve_error, search_errors, check_malicious, check_typosquat, get_scorecard, get_maintainer_trust, get_quality, get_provenance, get_trending, report_anomaly, contact_depscope, check_bulk, install_command, pin_safe.
Example response
{
"package": "express",
"ecosystem": "npm",
"latest_version": "5.2.1",
"health": {
"score": 85,
"risk": "low",
"breakdown": {
"maintenance": 25,
"security": 25,
"popularity": 20,
"maturity": 15,
"community": 10
}
},
"vulnerabilities": { "count": 0 },
"recommendation": {
"action": "safe_to_use",
"summary": "express@5.2.1 is safe to use (health: 85/100)"
}
}
Health score (algorithmic, 0-100)
Pure math, no LLM in the hot path. Runs in milliseconds.
| Signal | Max | Source |
|---|---|---|
| Maintenance | 25 | Days since last release. |
| Security | 25 | CVEs from OSV, filtered to the latest version. |
| Popularity | 20 | Weekly downloads from the registry. |
| Maturity | 15 | Total version count. |
| Community | 15 | Maintainers + GitHub stars. |
Key detail: we only surface vulnerabilities that actually affect the latest version. Django goes from 272 historical "vulnerabilities" to just the ones that still matter today.
Current average health across the 31k indexed packages: 60 / 100.
Self-hosting
DepScope is MIT. Everything you need is in this repo:
- API: FastAPI (Python 3.13) —
api/main.py+api/registries.py. - Frontend: Next.js 16 —
frontend/. - DB: PostgreSQL 17 with the schema in
api/database.py. - MCP: Node 20 —
mcp-server/. - Cron: 36 scheduled jobs listed in
CLAUDE.md§4. - Backups:
scripts/full_backup.sh— pg_dump + tarball + restic to S3.
Stage mirror runs side-by-side on different ports behind HTTP basic auth (see ecosystem.stage.config.js).
Ecosystem
- MCP Remote (29 tools, zero install):
https://mcp.depscope.dev/mcp - MCP Server (stdio):
depscope-mcpon npm - ChatGPT GPT: search "DepScope" in the GPT Store
- RapidAPI: hub listing
- OpenAPI Spec: depscope.dev/openapi.json
- AI Plugin Manifest: depscope.dev/.well-known/ai-plugin.json
- llms.txt: depscope.dev/llms.txt
Built with
FastAPI · PostgreSQL 17 · Redis · Next.js 16 · Node 20 · Python 3.13 · Proxmox 9.
Operated by Cuttalo srl. Feedback: depscope@cuttalo.com.
License
MIT — see LICENSE.
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
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
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.