How it works

1. Add the config below to your AI client
2. Your AI client opens a browser for authorization
3. Log in to PingZen (Google, Yandex, Telegram, or Email)
4. Click "Allow" — and MCP is connected!

Cline (via API Key)

{
  "mcpServers": {
    "pingzen": {
      "type": "streamableHttp",
      "url": "https://pingzen.dev/mcp/",
      "headers": {
        "Authorization": "Bearer pz_YOUR_API_KEY"
      }
    }
  }
}

Replace pz_YOUR_API_KEY with your actual key. Make sure to include "type": "streamableHttp" — without it, Cline defaults to SSE.

Available Tools

Tag Filtering

new

130 tools give full API coverage, but not every agent needs all of them at once. Tags let you get just the right group of tools with a single URL parameter:

• →Cursor (40-tool limit) — connect with relevant tags instead of all 130
• →Cline — fewer tools in context = better tool selection and lower token usage
• →CI/CD agents — connect only monitors,incidents for automation without clutter
• →Claude Code / Claude Desktop — works automatically (defer_loading), no tags needed

Available tags: monitors, alerts, channels, incidents, status_pages, heartbeats, workspaces, groups, telegram, reports, maintenance, webhooks, webhook_checks, web_push, dashboard, auth, feedback, probes, meta

Tool Search

new

The AI agent doesn't need to scan all 130 definitions — search_tools finds the right tool from a task description. The agent asks "what can the server do with SSL?" and gets a ranked list.

search_tools({ "query": "SSL certificate", "top_k": 5 })
→ create_monitor (score: 4.2), get_monitor_sla (score: 2.1), ...

Supports tag filtering — you can search only among alerts or only among monitors.

Claude Code uses search_tools automatically via defer_loading — all 130 tools are available without context overflow. No configuration needed.

Example Prompts

• •"List all my monitors"
• •"Create an HTTP monitor for example.com"
• •"What's the uptime for my sites?"
• •"Set up a Telegram alert for my monitor"
• •"Show me open incidents"

Troubleshooting

Client Support

Code Mode

experimental

Instead of calling tools one by one, the AI agent writes a Python script that executes multiple calls at once in a sandbox. Intermediate data (list of 300 monitors, report) stays in the sandbox and doesn't pollute the context window.

search("monitor uptime") → create_monitor, get_uptime_stats, ...
get_schema(["create_monitor"]) → { parameters: { url, protocol, ... } }
execute("""
  monitors = await call_tool("list_monitors", {})
  down = [m for m in monitors if m["status"] == "down"]
  return f"{len(down)} monitors down"
""")

Uses FastMCP 3.2 + Monty sandbox (Rust, 15s timeout, 50MB memory). Separate endpoint — the main /mcp/ works as before.