Crawl the target
Fetch the submitted URL, robots policy, sitemaps and prioritized internal pages. This establishes what the target says about itself.
URL Intelligence Agent
LIVE DOCKER SPACE · WEB-WIDE EVIDENCE · REMOTE MCP · MIT
URL in.
URL in.
Evidence out.
Evidence-first URL intelligence for AI agents, developers and research workflows. Full investigation starts with the target website, then expands beyond it into eligible third-party links, structured-data references and search-index results so external articles and independent-domain evidence can be checked instead of treating one website as its own proof.
First-party crawlThird-party evidenceBacklink-style discoverySource provenanceContradictionsRemote MCPTypeScriptDockerRAGMIT
Try a live investigation
Sign in with Hugging Face, enter a public URL and select an action. Full investigation performs the broadest research pass. The hosted demo allows one analysis every 24 hours per normal account; the owner account is exempt. Successful results can be exported as branded PDF, JSON, Markdown or HTML.
External index: Google Custom Search is preferred for the hosted Space. Availability is detected from server-side configuration; unavailable providers are disabled automatically.
🔐Hugging Face login is required. Sign in to use hosted analysis and report exports. Local/self-hosted usage remains unrestricted under the MIT License.
Collecting first-party evidence, expanding to third-party sources and checking contradictions… this may take a little time.
Complete raw response · nothing hidden
How full investigation researches the web
A site cannot independently confirm its own claims. The full workflow therefore separates first-party extraction from external corroboration and records what was actually checked.
Resolve identity
Use JSON-LD, Open Graph, titles, content and canonical signals to identify the entity and retain the exact first-party source URLs.
Discover every eligible external lead
Collect outbound links, public social/profile URLs and URLs embedded in structured data. Direct references from the target receive their own high-priority verification lane, while the selected web index adds independent articles and backlink-style results.
Open, read and verify the destination
The agent does not increase confidence just because the target published a link. It fetches eligible external articles, websites and public social/profile pages through the SSRF-safe network layer, parses the returned content, checks for entity mentions and verifies whether the source links back to the target. If a platform blocks public automated access, that source remains visible but unverified.
Score coverage separately
Extraction confidence and external corroboration are separate. Repeating the same claim across 30 pages on one domain does not become 30 independent sources.
Keep uncertainty visible
Source URLs, observation time, search-provider status, coverage limitations and explicit contradictions remain visible in the result and exports.
ℹ️External links are leads, not proof: eligible external URLs found on the target are opened and content-checked before they can strengthen corroboration. Articles are read and parsed; public social/profile pages are also attempted directly. A crawler still cannot enumerate the entire internet by itself, so broad backlink/article discovery additionally uses the selected search index.
Built inside HORNO Network. Now open source.
URL Intelligence Agent was not created as a standalone demo. It was developed, tested and refined inside the HORNO Network ecosystem, where its URL intelligence, evidence collection and enrichment workflows run in real production use cases. After proving the technology in a live ecosystem, HORNO Network founder Vincenzo Picciuolo chose to release the project as open source so developers, AI builders, researchers and companies can inspect it, self-host it, extend it and build with it. The project is developed by Vincenzo Picciuolo / HRN Innovation Technologies Ltd.
URL Metadata & Social Profile Fetcher
The lightweight deterministic companion for fast URL enrichment, Open Graph and SEO metadata, canonical URLs, images and public social/profile discovery.
Explore URL Metadata & Social Profile Fetcher on GitHub ↗HORNO Network
The production ecosystem where URL Intelligence Agent has been developed, tested and used across live URL, identity, discovery and intelligence workflows.
Explore the HORNO Network ecosystem at horno.net ↗HORNO Space
HORNO Space is the digital identity and smart-link product connected to the URL enrichment and public-profile intelligence layer.
Explore HORNO Space digital identity at space.horno.net ↗Follow Vincenzo Picciuolo
Follow the founder for URL Intelligence Agent releases, open-source development and updates from the HORNO Network ecosystem.
Complete action catalog
The runtime exposes 34 actions. The hosted Space enables a controlled subset for public testing; the complete set is available through CLI, self-hosted HTTP API and local MCP.
Connect your AI with Remote MCP
The Space exposes a public Streamable HTTP MCP endpoint. Compatible clients can discover the hosted tool subset and call URL intelligence directly from an AI conversation.
Public Remote MCP endpoint
https://vpicciuolo-url-intelligence-agent.hf.space/mcpTransport: Streamable HTTP · Public read-only demo tools · No API key required for connection · One analysis tool call per MCP session every 24 hours. Tool discovery, initialization and health checks do not consume the allowance.
ChatGPT
Where custom Remote MCP apps/connectors are available, create a custom app, paste the endpoint, choose no authentication for this public demo, discover tools and ask ChatGPT to investigate a public URL.
Claude / other MCP clients
Add the URL as a remote Streamable HTTP MCP server. After initialization the client can discover the hosted public tools and call an analysis.
Protocol test
curl -i https://vpicciuolo-url-intelligence-agent.hf.space/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"test-client","version":"1.0"}}}'Frequently asked questions
The interface deliberately distinguishes what the agent extracted, what third-party sources corroborated and what remains uncertain.
Why are there now two confidence concepts?
Extraction confidence is about the reliability of extracting a field from the target's observable metadata/content. External corroboration is about how much supporting evidence was actually fetched across third-party domains. They answer different questions and should not be collapsed into one fake probability.
Does full investigation search outside the submitted domain?
Yes. It first crawls the target, then evaluates eligible external URLs discovered in links and structured data. If a search provider is configured, it also queries the wider web for external references and articles before fetching selected results.
Can the agent discover every backlink?
No. No normal crawler can know every page on the internet that links to a domain. A search index is required for broad backlink-style discovery, and every provider has its own index coverage. The report therefore records the provider and coverage limitations instead of claiming completeness.
What happens when sources disagree?
Explicit contradictions remain in the result and are surfaced before the detailed sections. The agent does not silently force a single consensus answer.
Does “no contradiction found” mean all sources agree?
No. It only means the current run did not detect an explicit contradiction. Absence of a detected conflict is not proof of consensus.
Which external search index does the hosted demo use?
Google Custom Search is the preferred hosted index. The selector beside the analysis type lists every supported provider and disables providers that are not configured. DuckDuckGo is available as the built-in fallback. The selected provider is recorded in the result and exported report.
Can I inspect every source?
Yes. Evidence-linked fields retain source URLs, external research lists discovered/fetched third-party pages, and the complete raw JSON remains available.
Are trust, SEO, security or compliance scores definitive?
No. They are explainable observations of public signals. Security audit is not penetration testing, compliance signals are not legal advice, and trust scoring is not a guarantee about a person or organization.
What URLs can be investigated?
Public HTTP/HTTPS URLs. The runtime validates destinations, blocks localhost/private networks and does not intentionally bypass authentication, CAPTCHAs or access controls.
Why does the hosted demo have a limit?
The public Space is for testing. A normal signed-in Hugging Face account gets one hosted analysis every 24 hours. Clone or self-host the MIT project for unrestricted usage under your own infrastructure limits.
Run the full agent locally
Clone the MIT-licensed repository for unrestricted development, automation and self-hosting.
Quick start
git clone https://github.com/vpicciuolo/url-intelligence-agent.git cd url-intelligence-agent npm install npm run build node dist/src/cli.js investigate https://example.com node dist/src/cli.js seo https://example.com node dist/src/cli.js security https://example.com node dist/src/cli.js socials https://example.com node dist/src/cli.js technologies https://example.com
HTTP API, Docker and local MCP
Use the same runtime as a service or integrate it into other agent stacks.
HTTP API
npm run serve
Endpoints include /health, /actions, /investigate and /action/:name.
Local MCP
npm run mcp
Starts the stdio MCP server with the complete action registry for compatible local AI clients.
Docker
Build and run the same Node.js runtime as a portable container for private infrastructure or production integrations.
Open source, inspectable and reusable
URL Intelligence Agent is available under the MIT License. Inspect the source, run it locally, self-host it, integrate it into an AI workflow or contribute improvements on GitHub.