Skip to main content

isinjected

Is this prompt malicious?

A trust gate for AI agents. Scan any fetched page, email, ticket, or tool result for prompt injection — and check every link it contains against 30M threat indicators — before your agent acts on it.

Start freeRead the API docs

1,000 scans/month free, no credit card. Scans are metered separately from your API request quota.

Instruction override & jailbreaks

"Ignore all previous instructions", fake system prompts, role hijacks, and DAN-style jailbreaks — caught before they reach your model.

Invisible & encoded payloads

Unicode tag-smuggling, zero-width characters, bidi overrides, and base64/percent-encoded instructions are decoded and flagged.

Data exfiltration & malicious links

Markdown-image beacons, data: URIs, and any URL/domain/IP checked against known-malicious infrastructure.

Tool-call forgery

Forged tool calls, function-call structures, and shell exhortations that try to drive your agent's tools.

One call before your agent acts

Send the untrusted content; get back block, warn, or allow, with the flagged spans and a sanitized copy.

curl -X POST "https://ismalicious.com/api/gate/scan" \
  -H "X-API-KEY: $ISMALICIOUS_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content": "…fetched page text…", "source_url": "https://example.com"}'

Marketing.injectionScanner.mcpNote

Add it to your agent in one line

The isMalicious MCP server gives any MCP client — Claude Desktop, Cursor, Windsurf, your own agent — six tools over the same API, with typed errors an agent can act on.

npx -y @ismalicious/mcp-server

Or declare it in the client's MCP configuration:

{
  "mcpServers": {
    "ismalicious": {
      "command": "npx",
      "args": ["-y", "@ismalicious/mcp-server"],
      "env": {
        "ISMALICIOUS_API_KEY": "your-api-key",
        "ISMALICIOUS_API_SECRET": "your-api-secret"
      }
    }
  }
}

No key yet? Start without the two variables: the server offers a single bootstrap_key tool that mints a free key from an email address, one per IP per day.

  • scan_before_usePrompt-injection scan plus link reputation over untrusted content; block, warn or allow.
  • check_urlReputation of one URL, domain or IP before fetching it.
  • check_indicatorFull verdict for an IP, domain, URL or hash: risk score, citing blocklists, timeline, network, a headline you can relay and a recommended action.
  • get_cveOne CVE with CVSS, EPSS, CISA KEV status and due date, exploitation evidence and references.
  • recent_cvesLatest CVEs, optionally by severity.
  • bootstrap_keyWithout a configured key: mint a free one from an email address.

npm · MCP registry: com.ismalicious/mcp-server