MCP server

Orbilo for AI assistants

Connect ChatGPT, Claude, Cursor, VS Code, Codex or Claude Code to Orbilo and ask it whether a site is ready for agents. The checker works with no account at all. Sign in once and the same assistant can watch the sites you monitor.

POST https://orbilo.co/mcp

OAuth 2.1 with PKCE
Streamable HTTP
5 tools, 2 resources
Checker needs no account

Connect your assistant Check a site in the browser instead

What you can ask

The question "does this site work for agents" comes up in the editor, in a client thread and in a planning doc. With Orbilo connected, the assistant runs the check where the question was asked and answers from the result.

“Can AI agents use shop.example.com?”

A score and grade for UCP, WebMCP and Access, every check that passed or failed, and a fix for each. No account needed.

Calls check_site

“Does our checkout page register WebMCP tools properly?”

Pass the page and spec: webmcp. WebMCP is page-specific, so the assistant checks the page that declares the tools.

Calls check_site

“Here is a report link a client sent me. What should they fix first?”

Reads the shared report instead of checking the site again, and orders the fixes by impact.

Calls get_report

“How are the sites we monitor doing?”

The latest grade per spec on every site, open issue counts and how many site slots are left.

Calls list_monitored_sites

“What is wrong with app.example.com and when was it last checked?”

Every open issue with its fix, from the stored checks rather than a fresh run.

Calls get_site_status

“Start monitoring the new marketing site.”

Checks it now, then daily, and alerts the team when readiness changes. Uses one site slot.

Calls add_monitored_site

“I just added UCP support. Check again, skip the cache.”

fresh: true bypasses the 10 minute result cache so the assistant sees the deploy you just made.

Calls check_site

Connect in a minute

Every client needs one thing: the URL. Checking a site works immediately. The first time you ask for anything about monitored sites, the client opens the Orbilo consent screen; approve it once and the assistant gets a token limited to your team.

Claude Code

claude mcp add -s user --transport http orbilo https://orbilo.co/mcp

The -s user flag registers the server for every project. Checking a site works straight away; the first monitoring call opens the consent screen.

Codex

codex mcp add orbilo --url https://orbilo.co/mcp

Run codex mcp login orbilo when you want the monitoring tools.

Cursor

Add to .cursor/mcp.json:

{
    "mcpServers": {
        "orbilo": {
            "url": "https://orbilo.co/mcp"
        }
    }
}

Or open Cursor Settings, then MCP, then Add new MCP server, choose Streamable HTTP and paste the URL.

VS Code

Add to .vscode/mcp.json:

{
    "servers": {
        "orbilo": {
            "type": "http",
            "url": "https://orbilo.co/mcp"
        }
    }
}

Or run MCP: Add Server from the command palette, choose HTTP and paste the URL.

ChatGPT

Open Settings, then Connectors, then Create. Developer mode has to be on under Advanced settings.
Name it Orbilo, paste the server URL and save. Leave authentication on OAuth.
Ask it to check a site. It only asks you to sign in when you touch monitoring.

Server URL: https://orbilo.co/mcp

Claude

Open Settings, then Connectors, then Add custom connector.
Name it Orbilo and paste the server URL. Leave the client ID and secret empty.
Click Add. Connect when you want the monitoring tools; the checker works without it.

Server URL: https://orbilo.co/mcp

Works on claude.ai and in Claude Desktop.

Gemini CLI

gemini mcp add --transport http orbilo https://orbilo.co/mcp

Windsurf

Add to mcp_config.json:

{
    "mcpServers": {
        "orbilo": {
            "url": "https://orbilo.co/mcp"
        }
    }
}

Any other client that speaks Streamable HTTP works with the same URL. Full setup notes and troubleshooting are in the MCP docs.

Tools

This list is read from the running server, so it is exactly what a connected assistant sees. Every tool carries hints that tell the assistant whether it only reads or changes something, and whether it needs an account.

check_site

Check a Site No account needed Read only Live check

Check whether a website is ready for AI agents against the UCP, WebMCP and Access specs. Use this when someone asks "can agents use my site", whether a site supports UCP or WebMCP, or why an agent cannot buy from or act on a page. Returns a score and grade per spec, every check that passed, warned or failed with a fix for each, and a shareable report link. Runs live and can take up to a minute for all three specs. Works without signing in; signed-in users get a higher daily limit.

Every plan, daily check limit

url*: The page to check, e.g. https://shop.example.com. A bare domain is checked over https.
spec: Which spec to check: ucp, webmcp, access, or all (default). WebMCP checks are page-specific, so pass the page that registers tools.
fresh: Skip the 10 minute result cache and check the site again now.

get_report

Get a Shared Report No account needed Read only

Fetch a check report by its share token, the 16 character code at the end of an orbilo.co/check/r/... link. Use this when someone pastes a report link or asks about a check that was already run, instead of checking the site again. Reports expire 30 days after the check. Works without signing in.

Every plan

token*: The share token, or the full report URL

list_monitored_sites

List Monitored Sites Needs an account Read only

List the sites your Orbilo team monitors, with the latest grade per spec (UCP, WebMCP, Access), how many issues are open on each, and how many sites the plan still allows. Use this when someone asks what they are monitoring, how their sites are doing, or whether they can add another site. Requires a connected Orbilo account.

Every plan

No inputs.

get_site_status

Get Site Status Needs an account Read only

The current standing of one monitored site: the latest stored check per spec (UCP and Access for the site, WebMCP per monitored page) and every open issue with its fix. Use this when someone asks what is wrong with a site they monitor, what to fix first, or when it was last checked. Pass the domain or the site id from list_monitored_sites. For a site that is not monitored, use check_site instead. Requires a connected Orbilo account.

Every plan

site*: The site domain (e.g. shop.example.com) or its id

add_monitored_site

Add a Monitored Site Needs an account Writes

Start monitoring a site: Orbilo checks it against UCP, WebMCP and Access now and on a schedule, and alerts the team when readiness changes. Use this when someone wants ongoing monitoring rather than a one-off check, or asks to be told when their site breaks for agents. Uses one of the plan's site slots. Requires a connected Orbilo account.

Uses a site slot: 1 on Free, 5 on Pro, 25 on Scale

url*: The site to monitor, as a domain or URL, e.g. shop.example.com. Monitoring is per site; WebMCP pages can be added on the dashboard.

Resources

Background the assistant can read before it advises. Both are the same plain-English release notes as the spec changelog.

UCP Changelog

orbilo://changelog/ucp

The latest UCP (Universal Commerce Protocol) spec releases explained in plain English: what changed, whether it is breaking, and what a site needs to do. Read this before advising on UCP changes.

WebMCP Changelog

orbilo://changelog/webmcp

The latest WebMCP spec releases explained in plain English: what changed, whether it is breaking, and what a site needs to do. Read this before advising on WebMCP changes.

What it will and will not do

The checker is public

check_site and get_report work without a token, exactly like the checker on this site. Results are cached for 10 minutes and a daily limit applies per address; signing in raises it.

Monitoring needs OAuth

The endpoint refuses API tokens. You approve an assistant once on a consent screen and it gets a token limited to your team. Access tokens expire after an hour and refresh silently for up to 90 days.

Revoke in one click

Every connected assistant is listed under Connected assistants on the API tokens page. Disconnect revokes its tokens immediately.

Nothing is deleted by it

There is no remove tool. An assistant can add a monitored site but cannot delete one, change your plan, touch billing or invite anyone. Those stay on the dashboard.

Plan limits still apply

add_monitored_site uses one of your site slots, the same as adding a site on the dashboard. When the slots are full the tool says so instead of adding it.

Rate limited

The server has its own limiter, so a runaway agent loop cannot hammer a site with checks or burn through your daily allowance in one go.

Setup guide and troubleshooting Launch post OAuth resource metadata Privacy Terms

Monitor one site free, from any assistant

Monitoring one site is free permanently, not as a trial. Create an account, connect your assistant, and ask it to watch your site.

Start free See pricing