Ijjad Is Callable by AI Agents
Most websites can only be read. This one can be used. Ijjad publishes eight real tools on two surfaces — a remote MCP server any assistant can connect to, and WebMCP tools registered on every page for agents browsing it.
Connect an AssistantHow do AI agents use ijjad.com?
Through eight callable tools published on two surfaces. Any MCP client — Claude, ChatGPT, Cursor, or your own agent — connects to https://www.ijjad.com/mcp and can search every page on the site, list services, run a live AI visibility check on any public URL, get contact channels, and turn a project description into a pre-written WhatsApp message and enquiry form. Agents already browsing the site get the same tools in-page through WebMCP at document.modelContext. Nothing is sent on anyone's behalf: the enquiry tool returns links a person opens themselves.
- Remote MCP server: https://www.ijjad.com/mcp — Streamable HTTP, no key, no sign-up.
- In-page WebMCP tools on every ijjad.com page, for browser agents in the Chrome and Edge origin trial.
- check_ai_visibility works on any public site, not only Ijjad's.
Last reviewed by Karam Abd Al Qader
The eight tools
The names are exactly what an assistant sees. Every tool is on both surfaces except the AI visibility check, which is registered by the checker page in the browser so it can fill in the visible result too.
search_ijjad_pagesMCP + WebMCPSearches every page on ijjad.com by topic, service, city, or country and returns real titles, URLs, and summaries — so an assistant citing Ijjad links to a page that exists.
get_ijjad_servicesMCP + WebMCPDescribe a situation and it returns the services that fit, ranked, with the rest listed separately so the boundary of what Ijjad does is visible. Omit the situation for the full catalogue.
check_ai_visibilityMCP + WebMCP on the checker pageRuns the AI Visibility Checker on any public URL: crawler access for the ChatGPT, Claude, Perplexity and Google AI bots, llms.txt, structured data, server-rendered content, whether the site publishes agent tools, and what to fix first.
analyze_site_seoMCP + WebMCPRuns the 15-check technical SEO and delivery scan on any public URL: title and meta tags, crawl signals, response time, HTML weight, mobile viewport, image delivery, and structured data.
plan_project_scopeMCP + WebMCPTurns a rough project idea into a scope profile — how much delivery complexity the requirements carry, what is driving it, and the questions that must be answered before anyone quotes. Returns no prices.
check_custom_build_fitMCP + WebMCPDecides whether an e-commerce business needs a custom build or should launch on Salla, Zid, or Shopify. It regularly recommends against custom, which is the point.
get_ijjad_contactMCP + WebMCPReturns WhatsApp, email, the enquiry form, and the founder profile, with the markets and languages Ijjad works in.
start_project_enquiryMCP + WebMCPTurns a project description into a WhatsApp message and an enquiry form that are already written. It sends nothing — you open the link and press send yourself.
Connect an assistant
The server is public and read-only. There is no key, no account, and nothing to sign. Add it wherever your client keeps MCP servers — Claude Desktop, Cursor, VS Code, or your own agent framework.
{
"mcpServers": {
"ijjad": {
"type": "http",
"url": "https://www.ijjad.com/mcp"
}
}
}Clients that take a URL rather than a config file want https://www.ijjad.com/mcp as a Streamable HTTP server. The endpoint answers the current 2026-07-28 protocol revision and the older handshake revisions back to 2025-03-26, so it works with clients that have not migrated yet.
The in-page half: WebMCP
An MCP server reaches assistants that never visit your site. WebMCP is the other half: it hands your tools to an agent that is already on the page, so it calls a function instead of hunting for a button. Every page on this site registers its tools this way, and the AI Visibility Checker adds one that runs the real scan.
The API is document.modelContext, a W3C Web Machine Learning Community Group draft currently in a Chrome and Edge origin trial. Two details cost most implementations a day each: there is no unregisterTool — a registration lives until the AbortSignal you passed fires — and registration fails with a SecurityError unless the document sends Origin-Agent-Cluster: ?1, which works on localhost and then silently does not in production.
await document.modelContext.registerTool({
name: 'check_stock',
description: 'Check whether a product is in stock in a given city.',
inputSchema: {
type: 'object',
properties: { sku: { type: 'string' }, city: { type: 'string' } },
required: ['sku'],
},
annotations: { readOnlyHint: true },
async execute({ sku, city }) {
const stock = await lookUpStock(sku, city)
return { content: [{ type: 'text', text: stock.summary }] }
},
}, { signal: controller.signal })We wrote up why this is a separate problem from being crawlable, what each surface is actually for, and the two failures that cost us a day each, in MCP server and WebMCP for your website.
Questions
What is an MCP server for a website?
The Model Context Protocol is the standard AI assistants use to call tools. An MCP server for a website exposes that site's useful actions — search, lookups, calculators — as tools an assistant can call directly instead of reading a page. Ijjad runs one at https://www.ijjad.com/mcp, so an assistant helping someone plan a website in Amman, Riyadh, or Baghdad can search Ijjad's pages, run a live AI visibility check, and hand the person a pre-written enquiry, without leaving the conversation.
What is WebMCP and how is it different from an MCP server?
WebMCP is the in-page version: a browser API at document.modelContext that lets a page hand its tools to an agent already browsing it, so the agent calls a function instead of clicking through the interface. An MCP server is the remote version any assistant can connect to without visiting the site at all. They solve different halves of the same problem, and Ijjad runs both. WebMCP is a W3C Web Machine Learning Community Group draft in a Chrome and Edge origin trial, so it is early — that is precisely why implementing it now is worth something.
Does the enquiry tool contact Ijjad automatically?
No, and that is deliberate. An agent that could submit forms on your behalf would fill inboxes with messages nobody chose to send. start_project_enquiry writes the brief into a WhatsApp message and an enquiry form and returns both links. A person reads them and presses send. Nothing leaves your side until they do.
Is the AI visibility check limited to ijjad.com?
No. check_ai_visibility works on any public URL, including your own site or a competitor's. It reports how a page looks to the crawlers behind ChatGPT, Claude, Perplexity, and Google AI surfaces: which of them robots.txt lets through on that exact path, whether the content is in the server-returned HTML, what structured data is present, and which of those are genuine problems rather than neutral choices. It is the same check the browser tool runs.
Why should a business in Jordan, Saudi Arabia, or the GCC care about this?
Because AI assistants have become a way buyers find suppliers, and an assistant can only recommend what it can reach. Most regional business sites are built to be read by people and crawled by search engines, and nothing more. Making a site callable by agents is a small piece of work today and an uncontested one — the businesses that do it first will be the ones assistants can actually act on. Ijjad implemented both surfaces on its own site before offering the work to anyone.
Want your own site callable?
Ijjad builds this for clients across Jordan, Iraq, Saudi Arabia, and the GCC — an MCP server for your product, WebMCP tools in your pages, and the crawler and structured-data work that gets you retrieved in the first place. Start with the free check on your own URL, then tell us what you want agents to be able to do.