Developer docs: https://docs.dugganusa.com (for AI: https://docs.dugganusa.com/llms.txt)
Threat intelligence you can check. I build and run DugganUSA, AI security you can check, built with AI as a working partner: the DugganUSA Feed (free STIX/TAXII/MISP, live counts), and research on MCP and AI-agent security. Every claim links to its source, confidence is capped at 95%, and mistakes are corrected in public.
Before this: lead architect for Dell EMC's Azure Stack hybrid cloud, then cloud security architecture at Check Point and Palo Alto Networks.
Keep the tools and vendors you already use. These sit beside them, read-only, and need no key, no signup and none of your data.
- Before your agent installs a package or calls an MCP server, give it our checks. Claude Code:
claude mcp add --transport http dugganusa https://analytics.dugganusa.com/api/v1/mcpAny other MCP client: point it athttps://analytics.dugganusa.com/api/v1/mcp. Thecheck-packagetool answers allow, review or block for npm, PyPI, RubyGems and GitHub Actions, with the reason and the source. - Before a pull request merges a new dependency or MCP server, add dugganusa-checks-action to the workflow. It runs without a key at anonymous limits.
- Before your agent obeys an instruction file, run
npx github:pduggusa/dugganusa-agent-guardin the repo to find hidden Unicode in CLAUDE.md, .cursorrules and AGENTS.md (source).
What each check covers, and what it asks of you.
| Check | Runs where | Covers | Needs a key? | What leaves your machine |
|---|---|---|---|---|
check-package (MCP tool) |
Your agent, via https://analytics.dugganusa.com/api/v1/mcp |
npm, PyPI, RubyGems, GitHub Actions; exact version adds an OSV vulnerability check | No | The package name and version you ask about |
Dredd check_mcp_server (MCP tool) |
Your agent, via https://analytics.dugganusa.com/api/v1/dredd/mcp |
MCP servers, including their full npm/PyPI dependency graph | No | The server name you ask about |
| dugganusa-checks-action | GitHub Actions, on pull requests | package-lock, npm-shrinkwrap, requirements*.txt, pyproject, poetry.lock, Gemfile(.lock), MCP config files, workflow uses: lines |
No (anonymous limits); a free key raises them | Package, server and action identities from the diff |
| dugganusa-agent-guard | Locally or in CI, zero dependencies | Hidden Unicode in CLAUDE.md, .cursorrules, AGENTS.md and similar files | No | Nothing |
| dugganusa-cli | Terminal, CI, or as a local MCP server | IPs, domains, hashes, CVEs, and known-bad packages in lockfiles | Yes, a free key | The indicators it extracts, never the file |
All of them are read-only: they give a verdict and change nothing. Anything that could not be checked is reported as unknown, never as clean.
Every verdict cites where it came from: OSV, abuse.ch, CISA, the researchers who found the thing, and our own hunting. If one is wrong, tell us and we will fix it in public.
- API key in 30 seconds: analytics.dugganusa.com/stix/register
- Feeds: STIX 2.1 / TAXII 2.1 / MISP, CSV blocklists for firewalls and SIEMs
- CLI:
npx dugganusa-cli· MCP server: listed in the official MCP Registry asio.github.pduggusa/dugganusa-threat-intel
| Repo | What it does |
|---|---|
| dredd-mcp | Pre-flight security check for MCP servers before an agent calls them |
| dugganusa-agent-guard | Finds invisible Unicode prompt injection in CLAUDE.md, .cursorrules, AGENTS.md |
| dugganusa-edge-shield | Cloudflare Worker that blocks known-bad traffic at the edge |
| dugganusa-sentinel | Microsoft Sentinel data connector for the free feed |
| dugganusa-action | GitHub Action that flags known-bad indicators in pull requests |
| dugganusa-vscode | VS Code extension: threat lookups inside your editor |
| dugganusa-ietf | IETF Hackathon contributions on agentic-attack vectors and MCP verification |



