Is an AI tool, agent or MCP server alive, and since when? Checks from the outside, with proof.
Maintain this server? Claim it
Call this server's tools from your browser. Nothing to install.
Requests go through mcptry because browsers can't call most MCP servers directly. Header values are sent to the server for that request only and are never stored or logged.
Tool list from our last check. Connect live to refresh it.
Check any public website or hosted MCP server once, right now, and get what AgentViva saw: works, partly works, does not work, or blocked (the site refused the check or asks bots not to open it, which does not mean it is down), with the evidence (HTTP status, where the redirects ended, the page title, or the MCP tools it lists). No API key needed. One check is a snapshot, not a verdict over time: for a tool AgentViva watches every day, lookup_tool gives its confirmed state and history. The same address is answered from a 5-minute copy. Private and local addresses are refused.
The address, for example notion.so or https://example.com/mcp.
page (a website, the default) or mcp (a hosted MCP server).
Pick your client. Values in capitals, like YOUR_TOKEN, are yours to fill in.
Run in a terminal
claude mcp add --transport http agentviva https://agentviva.dev/mcp
Check any public website or hosted MCP server once, right now, and get what AgentViva saw: works, partly works, does not work, or blocked (the site refused the check or asks bots not to open it, which does not mean it is down), with the evidence (HTTP status, where the redirects ended, the page title, or the MCP tools it lists). No API key needed. One check is a snapshot, not a verdict over time: for a tool AgentViva watches every day, lookup_tool gives its confirmed state and history. The same address is answered from a 5-minute copy. Private and local addresses are refused.
| Argument | Type | Description |
|---|---|---|
| url* | string | The address, for example notion.so or https://example.com/mcp. |
| type | string | page (a website, the default) or mcp (a hosted MCP server). |
{
"type": "object",
"required": [
"url"
],
"properties": {
"url": {
"type": "string",
"maxLength": 2048,
"minLength": 1,
"description": "The address, for example notion.so or https://example.com/mcp."
},
"type": {
"enum": [
"page",
"mcp"
],
"type": "string",
"description": "page (a website, the default) or mcp (a hosted MCP server)."
}
},
"additionalProperties": false
}Look up an AI tool AgentViva already watches publicly, by name ("Notion") or address ("notion.so"), and get its state (up, degraded, down, blocked or unknown), since when, and its public status page. No API key needed. It reads what AgentViva already knows; it does not check the address now.
| Argument | Type | Description |
|---|---|---|
| query* | string | A tool name or an address. |
{
"type": "object",
"required": [
"query"
],
"properties": {
"query": {
"type": "string",
"maxLength": 100,
"minLength": 2,
"description": "A tool name or an address."
}
},
"additionalProperties": false
}Tell AgentViva that a tool it watches publicly does not work for you or your user, by name ("Notion") or address ("notion.so"), with an optional short note of what happened. AgentViva checks it again right away (at most once an hour per tool) and answers with what that check saw. A report never changes the tool's state by itself: only AgentViva's own checks do, and several must agree. One report per network, tool and day is counted. No API key needed. For a tool AgentViva does not watch, use check_url instead.
| Argument | Type | Description |
|---|---|---|
| note | string | What happened, in a few words (optional). |
| query* | string | A tool name or an address. |
{
"type": "object",
"required": [
"query"
],
"properties": {
"note": {
"type": "string",
"maxLength": 200,
"description": "What happened, in a few words (optional)."
},
"query": {
"type": "string",
"maxLength": 300,
"minLength": 2,
"description": "A tool name or an address."
}
},
"additionalProperties": false
}We read every tool's name, description and inputs the way the model will, looking for instructions hidden from you. A clean result means no warning signs, not a guarantee. Scanned Nov 14, 2026.