MCP server

Connect Agent Ping to ChatGPT, Claude or your AI editor and ask about your production agents in plain language: what is failing, why a run went wrong, what an evaluation said. Investigation and the fix happen in one place.

It is a hosted MCP server. There is nothing to install: you paste one URL, approve access in your browser, and you are connected.

Connect

The server URL is:

https://agentping.io/mcp

Add it as a remote / HTTP MCP server in your client. The first time you use it, the client opens a browser to Agent Ping; sign in, review the permissions, and approve. No secrets to paste, and you can revoke access any time from Settings, Connected apps.

It works with any MCP-capable client that supports streamable HTTP and OAuth 2.1. Discovery documents live at https://agentping.io/.well-known/oauth-protected-resource/mcp and https://agentping.io/.well-known/oauth-authorization-server.

ChatGPT

Once Agent Ping is listed in the ChatGPT plugins directory, find it under Settings, Apps and choose Connect. Until then, or on a workspace that allows custom connectors, open Settings, Connectors, Create, give the connector a name and paste the server URL above. Leave authentication on OAuth. ChatGPT registers itself with Agent Ping, sends you to the sign-in page and asks you to approve access. Enable it in a chat and ask about your agents.

Claude (claude.ai and Claude Desktop)

Once Agent Ping is in the Claude connectors directory, open Settings, Connectors, find Agent Ping and choose Connect. To add it by hand, choose Add custom connector and paste the server URL above. Leave the client ID and secret empty; Claude registers itself. Click Connect, sign in to Agent Ping and approve access. The connector is then available in every conversation and in Claude Desktop.

Claude Code

claude mcp add --transport http agentping https://agentping.io/mcp
claude mcp login agentping

Or add it to your project's .mcp.json and run /mcp inside Claude Code to sign in:

{
  "mcpServers": {
    "agentping": {
      "type": "http",
      "url": "https://agentping.io/mcp"
    }
  }
}

Cursor

Add Agent Ping to Cursor

The link opens Cursor and offers to add the server; approve it, then click Sign in next to the server in Settings, MCP. To add it by hand, put this in .cursor/mcp.json in your project root (or ~/.cursor/mcp.json for every project):

{
  "mcpServers": {
    "agentping": {
      "url": "https://agentping.io/mcp"
    }
  }
}

VS Code

Add Agent Ping to VS Code

The link opens VS Code and asks you to confirm the server. VS Code prompts you to sign in the first time Copilot uses it. To add it by hand, put this in .vscode/mcp.json, or run code --add-mcp '{"name":"agentping","type":"http","url":"https://agentping.io/mcp"}':

{
  "servers": {
    "agentping": {
      "type": "http",
      "url": "https://agentping.io/mcp"
    }
  }
}

Codex CLI

codex mcp add agentping --url https://agentping.io/mcp
codex mcp login agentping

Or add it to ~/.codex/config.toml:

[mcp_servers.agentping]
url = "https://agentping.io/mcp"

Gemini CLI

gemini mcp add --transport http -s user agentping https://agentping.io/mcp

Or add it to ~/.gemini/settings.json. Gemini CLI discovers the sign-in flow from the first 401 response:

{
  "mcpServers": {
    "agentping": {
      "httpUrl": "https://agentping.io/mcp"
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json and choose Sign in on the server in the MCP panel:

{
  "mcpServers": {
    "agentping": {
      "serverUrl": "https://agentping.io/mcp"
    }
  }
}

Other clients

Cline, Roo Code, Zed and Goose all take a remote MCP server: add the URL above as an HTTP (streamable) server and approve in the browser. Client UIs change; treat each client's own MCP docs as the source of truth.

What you can ask

Once connected, the assistant has read tools (list_teams, list_agents, search_runs, get_run_events, get_run_evaluation) and, on paid plans, tools to acknowledge incidents, give evaluation feedback, and change your monitoring configuration.

A worked example

You: Why did invoicer fail last night?

The assistant calls list_teams to find your team, then search_runs with agent: invoicer, status: failed, and last night's date range. It finds three failed runs, picks the first, and calls get_run_evaluation and get_run_events on it.

Assistant: invoicer failed three times between 02:00 and 02:40. The evaluation says the goal was not achieved: the agent called the lookup_customer tool, got an empty result, and produced an invoice with a blank customer. The upstream lookup returned 200 with an empty body, so nothing errored; the run "succeeded" technically but the evaluation caught the bad output. Here is the run: https://agentping.io/runs/run_eu_...

That is detection to diagnosis without leaving the editor.

What the assistant can and cannot do

Access is honest and bounded:

  • View your agents, runs, and results (the read permission, on every plan).
  • Acknowledge and resolve incidents, and give evaluation feedback (the ops permission, paid plans).
  • Change your monitoring configuration: rules, channels, budgets, status pages (the manage permission, paid plans, admin role).

A connected editor can never change your billing, invite or remove members, create or revoke API keys, or delete your account or your data. Those stay in the dashboard, on purpose: money through an assistant is a fraud surface, and deleting data needs a human in a browser.

Deletions through the assistant require an explicit confirmation that names the exact thing being deleted, so a wrong guess cannot remove the wrong rule. And every change an assistant makes is recorded in your Activity log (account settings) with who, when, via which app, and from what to what.

Disconnect

Revoke the connection from Settings, Connected apps in Agent Ping, or remove the server in your client. Revoking takes effect immediately.