WebMCP - use AlertKick from the agent in your browser
This page is the reference. For a walkthrough of actually using it, see AlertKick in your browser’s agent.
WebMCP is a W3C draft that lets a web page register tools with the browser’s AI agent. The agent calls the tool in the page, in the user’s logged-in session, instead of reading the screen and clicking. AlertKick exposes two WebMCP surfaces:
- The dashboard (
https://<team>.alertkick.com) registers the same 35 tools as the hosted MCP connector once you are logged in: list servers, alerts, monitors and heartbeats; create HTTPS, DNS, TCP, domain-expiry and mail-posture (SPF/DMARC/blocklist) monitors; acknowledge and resolve alerts; raise and approve change records. - The free checkers at alertkick.com/tools
each register one read-only tool (
check_ssl_certificate,lookup_dns_records,check_domain_expiry,check_security_headers,check_website,check_ip_reputation,check_mail_recordsfor SPF, DMARC, DKIM and blocklist status, andgenerate_mail_recordson the email setup wizard, which writes the MX, SPF, DKIM and DMARC records for the providers you name). No account needed.
Supported browsers
Section titled “Supported browsers”| Agent | Status |
|---|---|
| ChatGPT desktop app (built-in browser) | Works out of the box |
| Chrome 151 | Works out of the box; no flag needed |
| Chrome 146 to 150 | May need chrome://flags/#enable-webmcp-testing |
| Other browsers | Nothing is registered; the site works as normal |
WebMCP needs HTTPS and a top-level tab. Both AlertKick surfaces meet that.
You do not have to guess whether your browser supports it. Open any free tool and read the line above the form:
- WebMCP detected, N tools registered - your browser speaks WebMCP and this page’s tools are registered, so an agent can call them.
- Agent-ready - the browser has no WebMCP support. The page still works normally; nothing is registered.
- WebMCP detected, 0 tools registered - the browser exposes the API but the tools did not register. Reload; if it persists, please tell us.
What it looks like
Section titled “What it looks like”On a free checker page, no login:
Check the TLS certificate on example.com and tell me when it expires.
Is example.com’s mail set up properly, and are its mail servers on any blacklist?
Generate SPF, DKIM and DMARC records for example.com; we use Google Workspace and SendGrid.
In the dashboard:
Add uptime, DNS and certificate-expiry monitors for example.com, checking every minute.
What is paging right now? Acknowledge the critical ones and open a change window for tonight 22:00 to 23:00 on the web servers.
Safety model
Section titled “Safety model”- Read tools (
list_*,get_*, the free checkers) carryreadOnlyHint: trueand run without a prompt. - Write tools (
create_*,acknowledge_alert,resolve_alert,pause_monitor,delete_*,approve_change, …) open an Allow your AI assistant to … dialog showing the exact parameters. Nothing happens until you click Allow. The agent is paused for the duration throughrequestUserInteraction(). - Tools are registered when you log in and removed when you log out. A tab that is not signed in exposes nothing.
- Every call goes through the normal
/api/v1endpoints with your session, so tenant scope, roles and plan limits apply exactly as they do when you click.
Tool list
Section titled “Tool list”The dashboard tool names and input schemas are identical to the hosted MCP connector; see the tool table there. The manifest is generated from the MCP server, so the two cannot drift.
For developers
Section titled “For developers”The WebMCP layer is open source (MIT):
github.com/alertkick/alertkick-webmcp.
It contains the generated tool manifest, the browser adapters, and a small
registerPageTool() helper for adding a read-only tool to any page. The
README covers testing with navigator.modelContextTesting in Chrome.
To check what a page has registered in Chrome with the flag enabled:
await navigator.modelContextTesting.getTools();await navigator.modelContextTesting.executeTool('list_alerts', { status: 'open' });