# AgentViva

> AgentViva watches AI tools, agents and assistants from the outside and answers one question per target: is
> it alive, and since when. Four check types (page, json, canary, and mcp for a hosted MCP server and its tools), a
> signed webhook and Telegram for alerts, and public status pages, badges and JSON that need no key.

Base URL: https://agentviva.dev — auth is `Authorization: Bearer av_live_...`, one key per team.
All times are Unix milliseconds. States: unknown, up, degraded, down, blocked. A state changes only after
several probes agree; `blocked` means the target refused the probe and is never treated as dead.

## Docs

- [Getting started](https://raw.githubusercontent.com/AgentViva/docs/main/GETTING-STARTED.md): how to get access, add what to
  watch (dashboard, catalog, API, MCP), get alerts and show the status, step by step; also at
  https://agentviva.dev/docs
- [API reference](https://raw.githubusercontent.com/AgentViva/docs/main/api.md): the full v1 contract — monitors, bulk
  upsert, the catalog (one link to a list of tools, synced daily), probes, daily history (`days=N`), the change
  feed, alert deliveries, team settings, claims, bulk pause/resume/delete, "Needs a decision" (tools down for N
  days, with keep, false alarm, stop, or a signed `remove_requested` webhook to your own site), and the public
  surfaces
- [OpenAPI 3.1 spec](https://agentviva.dev/openapi.json): the same contract, machine-readable (also
  [openapi.yaml](https://raw.githubusercontent.com/AgentViva/docs/main/openapi.yaml)); agent discovery files at
  https://agentviva.dev/.well-known/ard.json and https://agentviva.dev/mcp/server-card
- [MCP server](https://agentviva.dev/docs#agent): `https://agentviva.dev/mcp`, 16
  tools; sign in in the browser (OAuth 2.1, discovered from the `401`) or send the Bearer key; `/mcp/public` has
  `lookup_tool`, `check_url` and `report_problem`, no key
- [Changelog](https://raw.githubusercontent.com/AgentViva/docs/main/CHANGELOG.md): every change to the public API
- [Agent skill](https://raw.githubusercontent.com/AgentViva/docs/main/skills/agentviva/SKILL.md): when an agent should
  reach for this API

## Public surfaces (no key)

- [Status JSON](https://agentviva.dev/v1/public/aigregator/notion.json): `{ state, since, last_probe_at, latency_ms, reason, evidence }`;
  `evidence` is what the last check saw (HTTP status, where redirects ended, the page title, MCP tools)
- [Badge](https://agentviva.dev/badge/aigregator/notion.svg): SVG (the default is a 180x48 card), cached five minutes. Seven forms (`?style=dot|word|pill|card|seal|strip|panel`), three themes (`&theme=dark|light|auto`)
- [Status page](https://agentviva.dev/s/aigregator/notion): HTML, 90-day history, no JavaScript
- [Tool list](https://agentviva.dev/status): HTML, the public tools by first letter and the ones not working right
  now, no JavaScript
- [Pricing](https://agentviva.dev/pricing.md): plans and prices in Markdown
- [Search](https://agentviva.dev/v1/public/search?q=notion): the public tools AgentViva already watches, by name or
  address, no key; `https://agentviva.dev/mcp/public` answers the same over MCP (`lookup_tool`)
- [Check now](https://agentviva.dev/v1/public/check?url=v0.dev): one check of any public site (`type=page`) or MCP
  server (`type=mcp`), right now, with what we saw; no key, 10 a minute per IP, the same address cached 5 minutes
  (MCP: `check_url`). The same check is a form at https://agentviva.dev/check
- [Liveness feed](https://agentviva.dev/v1/public/tools.json): every public tool AgentViva watches, with its domain,
  state, since when, and status/badge/JSON links; for directories to match against their list, no key, hourly.
  What a directory gets: https://agentviva.dev/for-directories
- [The state of AI tools](https://agentviva.dev/research): how many public AI tools work, why the rest fail, how many
  stop working each month; the numbers as JSON at https://agentviva.dev/research.json
- Report a problem: `POST https://agentviva.dev/v1/public/report` with `{ "query": "notion.so", "note": "…" }`
  (MCP: `report_problem`); we check the tool again, a report never changes a state by itself

## Optional

- [Crawler policy](https://agentviva.dev/bot): what AgentVivaBot does, how it signs its requests (Web Bot Auth,
  keys at https://agentviva.dev/.well-known/http-message-signatures-directory), how it obeys robots.txt, and how to
  opt out
- [Security policy](https://raw.githubusercontent.com/AgentViva/docs/main/SECURITY.md): how to report a vulnerability