Add AgentArmorHQ in a minute.

Quickstart

Three steps. About a minute.

1. Get a free key

   Sign in at agentarmorhq.com/account. Your first key is created for you with the command below already filled in. Free includes 100 checks a month.

2. Paste one command

   Pick where you work. Replace YOUR_API_KEY with your key.

3. Ask your assistant

   Then try: check this package before installing

claude mcp add --transport http agentarmor "https://agentarmorhq.com/api/mcp?key=YOUR_API_KEY"

{
  "mcpServers": {
    "agentarmor": {
      "url": "https://agentarmorhq.com/api/mcp?key=YOUR_API_KEY"
    }
  }
}

curl -fsSL https://agentarmorhq.com/api/cli -o ~/agentarmor.mjs

{
  "mcpServers": {
    "agentarmor": {
      "command": "node",
      "args": [
        "/Users/you/agentarmor.mjs",
        "mcp"
      ],
      "env": {
        "AGENTARMOR_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

export AGENTARMOR_API_KEY=YOUR_API_KEY
curl -fsSL https://agentarmorhq.com/api/cli | node --input-type=module - check axios@1.14.1

name: AgentArmor
on: [pull_request]
permissions:
  contents: read
jobs:
  gate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@FULL_COMMIT_SHA
      - uses: agentarmorhq/action@bfa0ba334769825920c7c729d6eb3260866774e8 # v1.1.2
        with:
          org-token: ${{ secrets.AGENTARMOR_ORG_TOKEN }}

curl -H "Authorization: Bearer $AGENTARMOR_API_KEY" \
  "https://agentarmorhq.com/api/v1/check?target=npm:axios@1.14.1"

Check your whole setup: paste your MCP config on /check. Secrets are removed in your browser first.

Download the 2-page setup guide (PDF)

What the answers mean

Every check tells you whether to install it right now. The answer is Block, Review, No known issues or Ungraded. Block is the only one that says do not install.

Real examples

Actual output, generated Oct 24, 2026. Answers change through the day, so run a check for the current one.

AgentArmorHQ Check: axios@1.14.1 (npm:axios)
Decision: BLOCK — Reported compromised in threat research: version 1.14.1 was hijacked. Do not install.
Risk Grade: Not graded yet
Outside Threat Reports: 3 report(s): The State of Open Source Supply Chain Attacks [compromised] (affected versions: 1.14.1, 0.30.4); Malware in axios [malicious] (affected versions: 1.14.1, 0.30.4); Malicious code in axios (npm) [malicious] (affected versions: 0.30.4, 1.14.1)
Registry (npm): latest 1.20.0, first published 4438 days ago, 1 maintainer
Caution: Version 1.14.1 does not exist on npm
Hardened Command: # DO NOT INSTALL axios@1.14.1 — remove immediately if present:
npm uninstall axios
Interactive Check: https://agentarmorhq.com/check?t=npm%3Aaxios
Threat reports checked Oct 24, 2026. Our opinion, not a code audit or security guarantee.

AgentArmorHQ Check: imgdecode (npm:imgdecode)
Decision: BLOCK — Reported malicious in threat research: every version of this package should be treated as harmful. Do not install.
Risk Grade: Not graded yet
Outside Threat Reports: 4 report(s): Our code analysis flagged this version [malicious] (affected versions: 1.0.3); MALFEX npm Package 'imgdecode' Hides Shellcode Loader in Windows Native Addon [malicious] (affected versions: 1.0.0, 1.0.1, 1.0.2, 1.0.3); Malicious code in imgdecode (npm) [malicious] (affected versions: 1.0.1, 1.0.0, 1.0.2); Malware in imgdecode [malicious]
Registry (npm): latest 0.0.1-security, first published 1 day ago, 1 maintainer
Caution: New package (published 1 days ago)
Hardened Command: # DO NOT INSTALL imgdecode — remove immediately if present:
npm uninstall imgdecode
Interactive Check: https://agentarmorhq.com/check?t=npm%3Aimgdecode
Threat reports checked Oct 24, 2026. Our opinion, not a code audit or security guarantee.

AgentArmorHQ Check: reqeusts-helper (npm:reqeusts-helper)
Decision: BLOCK — This package does not exist on npm. AI assistants sometimes invent package names, and attackers register them later. Do not install; check the spelling.
Risk Grade: Not graded yet
Outside Threat Reports: 0 reports on file
Hardened Command: npm install --ignore-scripts --save-exact "reqeusts-helper"
Interactive Check: https://agentarmorhq.com/check?t=npm%3Areqeusts-helper
Threat reports checked Oct 24, 2026. Our opinion, not a code audit or security guarantee.

AgentArmorHQ Check: axios (npm:axios)
Decision: NO KNOWN ISSUES — The latest version, 1.20.0, is not affected. Versions 1.14.1, 0.30.4 were hijacked and reported; never install those. Pin 1.20.0 exactly.
Risk Grade: Not graded yet
Outside Threat Reports: 3 report(s): The State of Open Source Supply Chain Attacks [compromised] (affected versions: 1.14.1, 0.30.4); Malware in axios [malicious] (affected versions: 1.14.1, 0.30.4); Malicious code in axios (npm) [malicious] (affected versions: 0.30.4, 1.14.1)
Registry (npm): latest 1.20.0, first published 4438 days ago, 1 maintainer
Hardened Command: npm install --ignore-scripts --save-exact "axios@1.20.0"
Interactive Check: https://agentarmorhq.com/check?t=npm%3Aaxios
Threat reports checked Oct 24, 2026. Our opinion, not a code audit or security guarantee.

A full report tells you what a package can reach on your computer and what it does when it installs: $19 once, or 3 a month on Pro.

MCP tools

The MCP server lives at https://agentarmorhq.com/api/mcp. Your assistant picks the right tool on its own; you can also name one.

CLI

One file, no dependencies, Node 18 or newer. Run it straight from the web:

curl -fsSL https://agentarmorhq.com/api/cli | node --input-type=module - check <target>

Flags

Team policy file

Put .agentarmor.yml (or .agentarmor.json, same keys) in your repo to set your own rules. A name matches with or without npm: or pypi: and with or without a version. An allow entry never overrides a confirmed Block. Only an ignore entry can, and it needs a reason. Unknown keys and missing reasons stop the run with exit code 2, so a typo cannot quietly turn a rule off.

# .agentarmor.yml
block_grades: [D, F]        # fail on these grades too
allow:                      # never fail on a grade for these
  - "@modelcontextprotocol/server-filesystem"
ignore:                     # skip completely; a reason is required
  - name: some-package
    reason: We use our own patched copy. Tracked in #123.

Your organization policy in the pull request check

The setup guides for AgentArmor for Teams are in the Teams docs.

If your team uses AgentArmor for Teams, the same check can read your organization policy. Create a CI token in your organization, save it as a repository secret, and pass it as org-token. The check then reads .mcp.json, .cursor/mcp.json, .vscode/mcp.json, .claude/settings.json, .windsurf/, mcp_config.json and every skill folder in .claude/skills.

      - uses: agentarmorhq/action@bfa0ba334769825920c7c729d6eb3260866774e8 # v1.1.2
        with:
          org-token: ${{ secrets.AGENTARMOR_ORG_TOKEN }}
          mode: strict

• It fails the job for a blocked item, for a changed pinned version or changed skill files, and, with mode: strict, for anything the policy does not list.
• An item that is still under review gets a warning.
• An item AgentArmor reports as malicious fails even if your organization approved it. Nothing in the repository file clears it, not even an ignore entry.
• The repository file can only make the rules tighter, unless an admin lists the repository in Settings. A listed repository's allow list can allow tools that are not on the organization list. It can never clear a Blocked item.
• If the policy cannot be fetched, the job fails with exit code 2.
• The result is sent to your organization: repository, commit, counts and the names of the items that failed.

# .agentarmor.yml
org: acme                   # your organization's address
block:                      # always fail, whatever the org says
  - "https://mcp.example.net/*"
  - some-package
ignore:                     # clear one failure; a reason is required
  - name: some-package
    reason: We use our own patched copy. Tracked in #123.

Exit codes

API

One endpoint, /api/v1/check. It returns the same answer as the MCP tool, as readable text and as JSON.

Single check

curl -H "Authorization: Bearer $AGENTARMOR_API_KEY" \
  "https://agentarmorhq.com/api/v1/check?target=npm:axios@1.14.1"

Batch

POST up to 100 targets (each up to 300 characters, body up to 64 KB). Results come back in order; each counts as one check.

curl -X POST https://agentarmorhq.com/api/v1/check \
  -H "Authorization: Bearer $AGENTARMOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"targets":["npm:axios@1.14.1","pypi:requests","github:owner/repo"]}'

Authentication and limits

A key is required. Send Authorization: Bearer aa_live_... (or X-AgentArmor-Key). Create one at /account.

Every response carries X-Quota-Limit, X-Quota-Remaining and X-Quota-Reset (the next reset, UTC).

Data sources

We combine public security advisories, malware reports and package signals, updated through the day.

Recommended setups

Ready-to-copy mcp.json files built from well-graded connectors, with install scripts turned off and access scoped narrowly.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-filesystem@2025.7.1",
        "/path/to/your-project"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-github@2025.4.8"
      ],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_READ_ONLY_PAT}"
      }
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@playwright/mcp@latest",
        "--headless",
        "--isolated"
      ]
    }
  }
}

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-postgres@0.6.2",
        "postgresql://readonly_user:${DB_PASSWORD}@localhost:5432/analytics_db"
      ]
    },
    "memory": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-memory@2025.4.8"
      ]
    }
  }
}

{
  "mcpServers": {
    "fetch": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-fetch@2025.4.8"
      ]
    },
    "notes": {
      "command": "npx",
      "args": [
        "-y",
        "--ignore-scripts",
        "@modelcontextprotocol/server-filesystem@2025.7.1",
        "/path/to/research-notes-folder"
      ]
    }
  }
}