UPTIME FOR AI TOOLS
Know the moment an AI tool stops working.
AgentViva checks your tools, APIs and AI assistants from the outside, every day, and tells you — with proof — when one breaks.
Invite-only for now · sign in with Google · connect your AI agent (MCP)
RIGHT NOW
AI tools watched from the outside
- Working4,152
- Not working133
- Partly working246
- Blocked us240
About one in 13 listed tools is broken or partly broken today.
WORKS WITH
WebhooksREST APIMCPClaude CodeCursorCodexTelegram (soon)See it in under a minute
How it works
1. Paste a URL
A website, a JSON health endpoint, an AI chat endpoint, or an MCP server. That is all we need to start.
2. We check it from outside
Like a visitor would, on the schedule you pick. By default, one failed check changes nothing.
3. You get the answer
An alert on your webhook (Telegram coming soon), a public status page and badge, and an API.
Four kinds of check
Is my website up?
We open the page like a visitor and follow redirects. We catch parked domains, dead hosts, sites that blocked us, and quiet moves to a new domain.
Does my API answer correctly?
We call a health URL and check the JSON reply with your rules: a field equals a value, a number stays under a limit, a key is there.
Does my AI give real answers? (Pro)
We send a real question to your AI chat endpoint. A model reads the answer and decides if it is a real answer or an apology.
Does my MCP server work?
We connect to your MCP server the way an AI app does, start a session and list its tools. Name the tools that must be there, and we tell you when one goes missing.
We never cry wolf
By default, one failed check changes nothing. The same result has to repeat before we call a tool down or send an alert, and the record says how many checks agreed.
If a site refuses our checks, we say blocked, never down, and we send no alert.
- upanswered, and correctly
- degradedslow, partly broken, or moved
- downconfirmed gone
- blockedthey refused us, and we say so
EXAMPLE: HOW A CHANGE IS CONFIRMED
14:02:03Z check 1 http_404 held at up 15:02:11Z check 2 http_404 held at up 16:02:07Z check 3 http_404 confirmed: down alert sent to webhook at 16:02:08Z
Who it is for
You build a tool
Watch it from the outside and show a status badge on your site or README.
You run a directory
Check every listing through the API and give each one a public status JSON. aigregator.com uses it. The free plan watches up to 25 tools. For a longer list, write to support@agentviva.dev.
Your tool is listed somewhere
Prove you own the domain and get its alerts too.
Connect from your code or your AI agent (MCP)
curl -X POST https://agentviva.dev/v1/monitors -H "Authorization: Bearer av_live_..." -H "content-type: application/json" -d '{"type":"page","name":"Geondex","external_id":"geondex","target":{"url":"https://www.geondex.com"},"public":true,"public_slug":"geondex"}'
curl https://agentviva.dev/v1/changes?since=0 -H "Authorization: Bearer av_live_..."
curl https://agentviva.dev/v1/public/aigregator/geondex.jsonclaude mcp add --transport http agentviva https://agentviva.dev/mcp
Then type /mcp in Claude Code, choose agentviva, then Authenticate, and press Allow in your browser. On claude.ai: Settings, Connectors, Add custom connector, with the same address. In Cursor or Codex, add the same address in their MCP settings and sign in.
Step by step: How to use AgentViva → AI agents
claude plugin marketplace add AgentViva/docs claude plugin install agentviva@agentviva
Then restart Claude Code, type /mcp, choose agentviva, then Authenticate, sign in and press Allow in your browser. The plugin brings both the MCP connection and the skill that tells your agent when and how to call AgentViva.
Step by step: How to use AgentViva → AI agents
One HTTP call per monitor. No agent to install and no SDK. Start with How to use AgentViva, then read the API reference, the OpenAPI spec and the agent skill. Questions: support@agentviva.dev.
Questions
Is it free?
Yes, while early access lasts. We will publish plans and prices before anything is charged. See pricing.
How often do you check?
You choose. A website is checked once a day unless you pick a faster schedule, as often as every hour. JSON and AI checks run every 6 hours and MCP checks every hour unless you pick another, as often as every 5 minutes.
What does "blocked" mean?
Some sites refuse automated visitors. Then we cannot tell if the tool works. We record blocked, we never call it down, and we send no alert.
Will you hammer my site?
No. We load one page per check, read at most 2 MB, and say who we are in the user agent. Read the bot policy.
Do I need to install anything?
No. Paste a URL in the dashboard. Developers can use one HTTP call instead.
Watch your first tool today
Ask for an invite with your Google account.