# Pingwright

> Monitoring for developers. Pingwright checks HTTP, GraphQL, gRPC, MCP and TCP endpoints from the internet and, through an agent installed on any machine in a customer's network, internal services the internet cannot reach. It alerts on Slack, Microsoft Teams, Discord, PagerDuty, email or signed webhooks, and can be driven by AI coding agents over MCP, including from CI when a pull request is merged.

## Monitor inside a private network

- Install the Pingwright agent on a machine inside the network; it checks internal hosts and ports and connects out to Pingwright over 443 (no inbound ports).
- Each machine has its own revocable token; several agents in one location share the work.
- Agent installers are not published yet.

## Monitor new endpoints from CI

A GitHub Actions workflow can run Claude Code (anthropics/claude-code-action) when a pull request is merged, with the Pingwright MCP server configured through `--mcp-config` and a Pingwright API key, allowing `mcp__pingwright__list_checks` and `mcp__pingwright__create_check`, so new endpoints get checks automatically. The landing page at https://pingwright.app shows the full workflow.

## Connect an AI agent

- MCP server (streamable HTTP): https://mcp.pingwright.app/mcp
- Sign-in: OAuth 2.1 with PKCE and dynamic client registration. Give the client only the URL; it opens a browser for the user to sign in and allow it.
- Headless use (CI, scripts): an API key (`pw_…`) created in the console, sent as `Authorization: Bearer pw_…`. A key works in one organization only.
- Claude Code: `claude mcp add --transport http pingwright https://mcp.pingwright.app/mcp`

## What an agent can do over MCP

- Create, update, pause, resume and delete checks; read a check, its recent results and its incidents.
- List organizations and locations.
- Create, update, delete and test alert channels; read alert delivery history.
- Mute and unmute a check's alerts, for a deploy or maintenance.
- Create, update, delete and read public status pages.

An agent acts with its user's role in each organization.

## Checks

- Types: HTTP (method, headers, body, expected status codes, text the body must contain), GraphQL (query and variables; a response with errors fails), gRPC (standard health service), MCP (the server and the tools it must advertise), TCP (connection, optional TLS).
- Interval 30 s to 24 h. A location counts as failing after a set number of consecutive failures (default 2).
- Locations: Pingwright's public region (Europe West today), and private locations where an agent inside the customer's network checks internal services.
- A check whose locations have no agent online shows "No data", never "Up".
- Header and metadata values are write-only and encrypted at rest.

## Alerts

- Channels: Slack, email (to organization members), Discord, Microsoft Teams, PagerDuty (opens and resolves PagerDuty incidents), signed webhooks to any public https URL.
- Per channel: events (down, recovered, reminders while down), which checks, and a delay before alerting.
- Flapping checks send one notice; checks failing together arrive as one message.
- Pingwright does not send SMS or phone calls itself.

## Status pages

Public pages at https://pingwright.app/status/<slug> showing chosen checks under public names, with 90 days of daily uptime and recent incidents. Real check names and targets are never shown. Each page has a badge at /status/<slug>/badge.svg.

## Links

- Console (sign up, sign in): https://console.pingwright.app
- Landing page: https://pingwright.app