phishunt for AI agents

Connect Claude Desktop, Cursor, Zed, or any Model Context Protocol (MCP) client to phishunt's public phishing-domains feed. Free, CC0, no auth, no rate limits beyond Cloudflare's.

Model Context Protocol Agent Skills (RFC v0.2.0) API Catalog (RFC 9727) CC0 licensed

Agent in a hurry? Fetch https://phishunt.io/agent-setup/prompt.md and follow it - a self-contained setup doc that registers and verifies the MCP server without any of the steps below.

Option 1MCP server

Remote MCP server at https://mcp.phishunt.io/ over HTTP JSON-RPC 2.0 (protocolVersion 2025-11-25). Works with any SDK-based client.

Config for Claude Desktop / Cursor / Zed

// Add to your MCP client config
{
  "mcpServers": {
    "phishunt": {
      "url": "https://mcp.phishunt.io/"
    }
  }
}

Config for Claude Code

claude mcp add --transport http phishunt https://mcp.phishunt.io/

The --transport http flag is required - Claude Code defaults to stdio and would otherwise try to run the URL as a local command.

Available tools

check_domain

Is a given domain or URL substring flagged in the active phishunt feed? Returns matching entries with full metadata.

list_brand_phishings

List active phishing sites targeting a specific brand (e.g. microsoft, binance).

get_recent_detections

Delta sync: detections since a given ISO date. Useful for periodic blocklist refresh.

get_brand_metadata

Curated metadata for a brand: display name, category, primary domain, AI-authored characterisation of why it's targeted, and count of active phishings.

get_cert_metadata

Factual metadata for a TLS intermediate CA seen on phishing sites: operator, root, key type, related siblings, count of active phishings using the intermediate.

search_phishings

Free-text substring search across active phishing URLs, domains, and IP addresses. Min 3 chars.

analyze_url

Analyze any URL for phishing signals without contacting it: live URL-shape heuristics, phishunt's stored verdict if known, and apex history. Suspicious unknown domains are queued for full analysis.

analyze_url_deep

ACTIVE deep analysis: actually fetches the URL (HTTP, TLS cert, RDAP, GeoIP via SOCKS5) and re-scores it with the full 5-layer engine. Slow (5-15s), shared 50/day budget, single-flight - use only when analyze_url is inconclusive.

get_related_infrastructure

Find infrastructure and content overlap between a known indicator and other detections: shared IP, TLS certificate, nameservers, favicon/screenshot, redirect target, or naming pattern. Observed technical overlap, not an attribution claim.

get_campaigns

List possible campaigns / suspected clusters: indicators that share infrastructure or content signals, computed by a daily correlation job. Shared-infrastructure grouping, not an attribution claim.

get_campaign

Full detail on one possible campaign: evidence breakdown and every member indicator (domain, targeted brand, status, relationship score, detail page).

Try it

curl -sX POST https://mcp.phishunt.io/ \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

Source on GitHub Service card

Option 2Agent Skills

For agents that consume Agent Skills (RFC v0.2.0) instructions instead of MCP tools. Each skill is a markdown file explaining how to call a specific phishunt endpoint.

Skill	Purpose
`fetch-feed`	Download the full feed in TXT, JSON, or CSV.
`lookup-brand`	Query the API for sites targeting a brand.
`check-recent`	Pull detections since a date (pagination pattern included).
`brand-metadata`	Fetch curated metadata for a tracked brand (display name, category, active count).
`cert-metadata`	Fetch metadata for a TLS intermediate CA seen on phishing sites.
`search-phishings`	Free-text search across active phishing URLs, domains, and IPs.
`analyze-url`	Passively analyze a URL: stored verdict, live URL-shape heuristics, and apex history.
`related-infrastructure`	Find shared-infrastructure overlap (IP, cert, nameservers, favicon) with a known indicator.
`list-campaigns`	List possible campaigns / suspected clusters sharing infrastructure or content signals.
`campaign-detail`	Full evidence breakdown and member list for one possible campaign.
`analyze-url-deep`	ACTIVE deep analysis of a URL. MCP-only, shared 50/day budget.

Skills index (JSON)

Option 3Direct REST API

If your agent doesn't do MCP or Agent Skills, call the REST API directly. Public, CORS-enabled, CC0 data.

Endpoint	Purpose
`GET /api/`	Full API documentation.
`GET /openapi.yaml`	OpenAPI 3.1 spec.
`GET /api/v1/domains`	Query endpoint (filter by brand, date, format).
`GET /api/v1/analyze`	Passive URL analysis (stored verdict + live heuristics + history). Never contacts the target URL.
`GET /feed.txt` / `.json` / `.csv`	Bulk feed download.

ReferenceDiscovery endpoints

Agents crawling phishunt.io can programmatically discover these integrations.

Path	Format	Spec
`/.well-known/api-catalog`	linkset+json	RFC 9727
`/.well-known/agent-skills/index.json`	JSON	Agent Skills RFC v0.2.0
`/.well-known/mcp/server-card.json`	JSON	MCP SEP-1649 draft
`Link` HTTP headers	HTTP header	RFC 8288
`/llms.txt`	text/plain	llms.txt (llmstxt.org)
`/llms-full.txt`	text/plain	llms.txt (full variant)
`/AGENTS.md`	text/markdown	AGENTS.md convention

Every response from phishunt.io advertises Link: <...>; rel="mcp-server", rel="api-catalog", rel="service-desc", rel="service-doc", and rel="agent-skills" so a crawler only needs a single HEAD / to discover everything.

LicenseData & code

Data (feed contents, API responses, MCP tool outputs): CC0 1.0 Public Domain. No attribution required, no warranty. Attribution still appreciated.

Source code for the MCP server: 0xDanielLopez/phishunt-mcp (MIT).