Model Context Protocol

Your AI agent checks scams - on its own.

Scampilot as an MCP server: Claude, Cursor and friends see our scam detection as a tool. The agent decides on its own when to check a link, an email, or an SMS - and shows the verdict as an interactive card.

Connect your client For developers

9 tools5 resources5 promptsHTTP + stdio

{
  "mcpServers": {
    "scampilot": {
      "transport": "http",
      "url": "https://scampilot.de/mcp/scampilot",
      "headers": {
        "Authorization": "Bearer sk_live_…"
      }
    }
  }
}

What is MCP?

The open standard for AI tools.

The Model Context Protocol connects AI assistants to real services. Wire in Scampilot and your agent gains scam detection, curated knowledge, and ready-made prompts - with your token, in your language.

Tools

Actions the agent calls - scan, read reports.

Resources

Readable knowledge - signal glossary, guides, reports.

Prompts

Ready templates - reply safely, report to authority.

The complete surface

Nine tools, five resources, five prompts.

Tools

scan_text

Check arbitrary text (SMS, chat).

scan_url

Check a single URL against reputation.

scan_email

Raw email incl. forwarded chains.

scan_image

Analyse a screenshot via OCR.

scan_qr

Decode a QR code and check where it points.

scan_links

Extract links, follow redirects.

scan_phone

Check a phone number for scam patterns.

get_report

Re-read a previous report by id.

list_my_recent_reports

Your recent checks.

Resources

scampilot://signals/glossary

What every signal key means.

scampilot://verdicts/scale

Read verdict + confidence.

scampilot://intel/guides

Index of curated guides.

scampilot://intel/guides/{slug}

One guide in full.

scampilot://reports/{id}

A single report by id.

Prompts

explain_report_to_my_parent

Translate a report for a relative.

draft_safe_reply

Respond safely instead of engaging.

report_to_authority

Where and how to report.

triage_inbox

Rank several messages by risk.

compare_messages

Same scam campaign?

Interactive verdict card

Not just text - a card you can read.

In hosts that support MCP apps, every scan renders an embedded card: colour-coded verdict, confidence meter, the detected signals, and the recommended actions - themed to your client.

safe / warn / danger verdict, colour-coded
Refresh without re-scanning
Bilingual (EN/DE)

DANGERConfidence 94/100

Phishing attempt impersonating your bank.

credential_request highlookalike_brand high

Connected in three steps

Token, config, done.

Create a token

In Dashboard → API tokens, create a bearer token with the submissions:create and reports:read abilities.

Connect a client

Add the HTTP endpoint and token to your MCP client (Claude Desktop, Cursor, Continue.dev) - or run stdio locally.

Ask Claude

"Is this SMS a scam?" - the agent calls scan_text on its own and shows you the verdict card.

HTTP (Claude Desktop)

{
  "mcpServers": {
    "scampilot": {
      "transport": "http",
      "url": "https://scampilot.de/mcp/scampilot",
      "headers": {
        "Authorization": "Bearer sk_live_…"
      }
    }
  }
}

Local as stdio

# Local stdio transport:
php artisan mcp:start scampilot
# → stdin/stdout MCP channel