Docs · MCP

Connect your AI agent

Avospy ships a native MCP server. Connect it once and your AI agent can audit any website, set up monitoring, read live uptime and alerts, create heartbeats and status pages, and pull your weekly report — in plain language, without opening the dashboard.

Works with Claude Code, Cursor, VS Code, Windsurf, Claude Desktop and any MCP client
Setup takes about two minutes
Revoke access anytime

1. Create an access token

Sign in and open Settings → AI / MCP Access, then click New Token. Name it after the client that will use it — “Claude Desktop”, “Cursor”, “CI” — so you can tell them apart later.

  • Tokens start with avospy_mcp_ and are shown exactly once — copy the value right away.
  • Only a SHA-256 hash is stored, so a lost token cannot be recovered; create a new one instead.
  • A token carries the full access of your account. Issue one per client so you can revoke them separately.

2. Point your client at the endpoint

The server speaks Streamable HTTP and authenticates with a bearer token. That is all your client needs:

Endpointhttps://avospy.com/api/mcp
TransportStreamable HTTP
HeaderAuthorization: Bearer avospy_mcp_…
Claude Code (terminal)
claude mcp add --transport http avospy https://avospy.com/api/mcp \
  --header "Authorization: Bearer avospy_mcp_YOUR_TOKEN"
Cursor, Windsurf and other clients with a JSON config
{
  "mcpServers": {
    "avospy": {
      "url": "https://avospy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer avospy_mcp_YOUR_TOKEN"
      }
    }
  }
}
VS Code (.vscode/mcp.json)
{
  "servers": {
    "avospy": {
      "type": "http",
      "url": "https://avospy.com/api/mcp",
      "headers": {
        "Authorization": "Bearer avospy_mcp_YOUR_TOKEN"
      }
    }
  }
}

Claude Desktop's config file only starts local servers, so it reaches the endpoint through the mcp-remote bridge (needs Node.js). Put this in claude_desktop_config.json and restart the app:

Claude Desktop (claude_desktop_config.json)
{
  "mcpServers": {
    "avospy": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://avospy.com/api/mcp", "--header", "Authorization:${AUTH}"],
      "env": { "AUTH": "Bearer avospy_mcp_YOUR_TOKEN" }
    }
  }
}

Not sure whether the token works? Call the endpoint directly — a valid token answers with a JSON-RPC listing of the tools below, an invalid or revoked one returns 401 Unauthorized.

Check the connection
curl -i https://avospy.com/api/mcp \
  -H "Authorization: Bearer avospy_mcp_YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

The reply arrives as an “event: message” line with a JSON-RPC payload — that is normal for Streamable HTTP, not an error.

3. Just ask

Once the client is connected, the tools show up in the conversation. You do not call them by name — describe what you want:

“I just deployed myshop.cz — check it for anything broken.”
“Add myshop.cz to monitoring and give it a public status page.”
“Make sure my nightly backup runs — set up a heartbeat and add it to the crontab.”
“Why did I get a downtime alert last night?”
“Which broken links does the latest crawl show? Fix them in the repo.”
“Which of my sites had the worst uptime over the last 30 days?”
“Pause monitoring on staging.myshop.cz, we are redeploying it.”
“Tell me when rival.com changes the price on their pricing page.”

4. What the agent can do

19 tools, all scoped to your own account. Plan limits apply exactly as they do in the dashboard.

ToolWhat it doesArguments
Check & audit
check_websiteAudits any public URL on the spot — availability, speed, SSL, meta tags and technical SEO, with a score. No project needed.url, fresh?
get_site_healthEverything we know about one project in one call: SSL, DNS, technical SEO, meta tags, Lighthouse, Core Web Vitals and broken links.projectId
get_broken_linksBroken links and dead sitemap URLs from the latest crawl, each with the page it was found on.projectId, limit?
run_scanStarts a fresh crawl for broken links, or a sitemap check, right now.projectId, type?
Read monitoring data
list_projectsEvery project in your account with its latest uptime check.none
get_projectOne project in detail: latest check, SSL status and a 7-day uptime summary.projectId
get_uptimeDay-by-day uptime and response times for the last 1–90 days (7 by default).projectId, days?
list_alertsRecent alerts — downtime, SSL expiry, DNS changes, new subdomains, watchdog hits — optionally for one project or unread only.projectId?, unreadOnly?, limit?
mark_alerts_readMarks alerts as read once your agent has dealt with them.alertIds?
get_reportYour latest weekly AI report — generated on the spot if none exists yet.none
Set up & manage
add_projectAdds a website. Uptime, SSL, DNS, SEO and performance checks start automatically.name, url
configure_projectRename a project, change its URL, or pause and resume monitoring.projectId, name?, url?, active?
delete_projectRemoves a project and all of its history. Requires an explicit confirmation.projectId, confirm
create_heartbeatCreates a heartbeat for a cron job or worker and returns the ping URL to call when it runs.name, intervalMinutes, graceMinutes?
list_heartbeatsAll heartbeat monitors with their status, last ping and ping URL.none
add_watchdogWatches one element of a page (CSS selector, XPath or regex) and alerts when its content changes.projectId, name, selectorType, selector
list_watchdogsWatchdogs of a project with their current value and the last detected change.projectId
add_competitorTracks a competitor's site — tech stack, social profiles and content changes.projectId, name, url
create_status_pagePublishes a public status page for a project and returns its URL plus Markdown for an uptime badge.projectId, slug, title

5. Keeping the token safe

  • Treat the token like a password: keep it in an environment variable or a secret manager, never in a committed config file.
  • Revoking a token in Settings takes effect immediately — the next request is rejected.
  • Each token shows when it was last used, so an unused one is safe to delete.
  • The endpoint is HTTPS-only and every request without a valid token is answered with 401.

6. When something does not work

401 Unauthorized

The token is mistyped or revoked, the header is missing the “Bearer ” prefix, or the value does not start with avospy_mcp_.

The client hangs while connecting

It is probably trying the legacy SSE transport. This server speaks Streamable HTTP only — set the transport type to “http”.

“Project not found (or not owned by you).”

The projectId belongs to a different account, or it is a name instead of an id. Run list_projects first and reuse the id from there.

“Project limit reached”

You hit the project quota of your plan. Change the plan in Billing, or pause a project you no longer need.

Uptime data comes back empty

Monitoring has just started. The first checks arrive within a few minutes; daily statistics appear after the first full day.

Still stuck? Write to support@avospy.com and include the name of your MCP client.

Ready to connect your agent?

Create an account, add your first site, and issue a token — all in a couple of minutes.

Connect your AI agent (MCP) | Avospy