The next billion visitors are AI agents. Is your website ready?
Agentify converts any website or API into a governed set of agent-callable functions — one MCP endpoint that Claude, ChatGPT, and Copilot can discover, authenticate against, and use safely. Your data stays in the Kingdom.
$ node scripts/e2e.js ■ Booting demo-site, auth, gateway… [transformer] openapi.json → 4 tools (3 read / 1 write) ✓ discovery — RFC 9728 + RFC 8414 chain ✓ 401 bootstrap with WWW-Authenticate ✓ gated client registration ✓ token bound to tenant (aud check) ✓ MCP initialize · tools/list · tools/call ✓ order created via agent — "orderId":1001 ✓ read-scope token denied write tool 🎉 E2E PASSED — 14/14 checks
How it works
From website to agent-ready in three steps
No rebuild, no SDK lock-in, no component library. Your stack stays exactly as it is — we add the agent layer on top.
Place the file & verify
Drop the small Agentify Connect File onto your site — once, ever; you never update it. We verify you own the domain, then scan your site and compile its capabilities into described agent tools: search_site, create_order, and more.
Review & choose exposure
Analysis is a proposal — you decide: pick exactly which tools agents can see and the approval policy per tool. Nothing publishes before you approve it. Every tool sits behind OAuth 2.1 with scopes and audience-bound tokens.
Stay managed, continuously
One MCP endpoint makes you visible to every MCP-capable agent. Re-analyze on demand when your site changes, refresh content for free, and govern every tool from a single dashboard.
Platform
Security-first infrastructure, not a chatbot wrapper
Standalone auth layer
A full OAuth 2.1 authorization server — dynamic client registration, PKCE, introspection — isolated by design. Point it at us or swap in your own IdP.
Audience-bound tokens
Tokens are minted per tenant and rejected everywhere else — the confused-deputy attack dies at the gateway, verified in our test suite.
Safe execution engine
JSON-Schema validation, SSRF host allowlists, response envelopes against prompt injection, and a stable error taxonomy on every call.
Human-in-the-loop approvals
Flag any tool as approval-required — the agent gets a pending handle, you get a WhatsApp or dashboard prompt. Closed by default.
Agent-traffic analytics
See which agents call what, success rates, latencies, and denied scopes. Agent traffic becomes a first-class analytics channel next to web traffic.
Built-in discovery
Generated llms.txt, .well-known/mcp.json, and an opt-in registry put your business where agents actually look.
Who it's for
Built for the businesses agents will call first
SaaS & digital products
Your customers are already asking to drive your product from Claude and ChatGPT.
- OpenAPI in, MCP out — live in an afternoon
- Per-tool scopes map to your existing plans
- Ship "works with Claude/ChatGPT" this quarter
Operations & commerce
Logistics, distribution, booking, e-commerce — make your services orderable by agents.
- Write actions gated by approval queues
- Arabic-first customer-facing flows
- A new sales channel, measured like one
Regulated enterprises
Deploy the gateway and auth layer inside your own perimeter.
- On-prem / sovereign-cloud distribution
- Line-by-line audit trail, PDPL DPA
- Standards-pure: swap any OAuth 2.1 IdP
Connect an agent
Two lines in any MCP client
{
"mcpServers": {
"your-business": {
"url": "https://gw.agentify.example/mcp/your-tenant",
"transport": "http"
}
}
}
// Compliant clients bootstrap OAuth automatically via WWW-Authenticate →
// /.well-known/oauth-protected-resource — register, consent, call tools.
FAQ
Questions, answered honestly
How is this different from Kong or Cloudflare's MCP gateways?
They sell to global platform teams that already run their API infrastructure. We serve businesses in Saudi and the Gulf that have a website or API but no platform team — self-serve onboarding, Arabic-first flows, PDPL-conscious data handling, and regional hosting. Different buyer, different product.
Do I need to rebuild my website or install an SDK?
No. The gateway sits in front of your existing site or API. If you have an OpenAPI spec, transformation is automatic; if not, the wizard lets you define functions manually. The only things you add to your site are the small Connect File (once, for ownership verification) and two optional discovery files.
What stops an agent from doing something destructive?
Four layers: per-tool OAuth scopes (a read-only token simply cannot call write tools), human approval queues on sensitive actions, schema validation that strips unknown inputs, and rate limits with a per-tenant kill switch. Everything is closed by default.
Which AI agents are supported?
Anything that speaks MCP — Claude, ChatGPT, Microsoft Copilot, Cursor, and thousands of custom agents. MCP is the industry standard with first-class support across all major AI clients; an OpenAPI skin for non-MCP frameworks is on the roadmap.
Where does my data live?
Hosted tenants run in-Kingdom; Sovereign-tier customers run the gateway and auth layer entirely inside their own infrastructure. Agent requests are proxied — your origin credentials never leave the vault, and we never train on your data.
Be reachable when the agents arrive
Self-serve, end to end: sign up, point us at your website, and watch it become an agent channel — free tier included, no waitlist, no sales call.
Get started — self-serve