MCP Server

Steuere PushWard aus Claude und anderen KI-Agenten über das Model Context Protocol. Der gehostete MCP Server lässt einen Agenten Live-Aktivitäten erstellen, Benachrichtigungen senden und Widgets über deinen Integrationsschlüssel verwalten.

Connect from Claude Code

Add the server with one command. OAuth runs interactively the first time the agent uses a tool.

claude mcp add --transport http pushward https://mcp.pushward.app/mcp

Or install the PushWard plugin, which adds the same server together with the agent skill below, then run /mcp and sign in:

/plugin marketplace add mac-lucky/pushward-mcp
/plugin install pushward@pushward

Connect from other clients

Any MCP client that supports remote servers over OAuth — Claude Desktop, the Claude apps, and others — can connect by adding a remote (custom) connector pointed at the endpoint below:

https://mcp.pushward.app/mcp

The client discovers the OAuth endpoints automatically and walks you through authorization on first connect.

Authorize

The first time you connect, a browser opens a PushWard consent screen. Paste your hlk_ integration key once and approve access. The server validates the key, stores it encrypted, and issues short-lived tokens for the session — your raw key is never shared with the MCP client.

Get your integration key from the iOS app's settings screen. See Authentication for how keys and their permissions work.

Agent skill

The tools tell an agent what it can call; the pushward-mcp skill tells it when. With it, "ping me when the migration finishes" becomes one notification at the end instead of one per step, an approval waits on wait_for_answer instead of a polling loop, every Live Activity it starts gets ended, failures included, and it leaves the integration key tools alone unless you ask. The plugin above includes it; for any other agent that reads SKILL.md files:

npx skills add mac-lucky/pushward-mcp --skill pushward-mcp

Agents that run shell commands rather than MCP tools have a twin skill for the CLI.

What you can do

The remote server exposes tools covering the full API surface:

• Live Activities — create, update, end, get, list, and delete activities, plus bulk-end by filter.
• Notifications — send inbox notifications with optional push delivery, schedule them once or on a repeat, and read answers.
• Answers — wait_for_answer blocks until the user answers an approval Live Activity or a notification with url-less actions.
• Widgets — full CRUD for Home Screen widgets.
• Email — send transactional email to verified recipients.
• Health — liveness and readiness checks for the API.
• Test workflows — composite helpers that run a full lifecycle (for example, create an activity, push a few updates, then end it) in one call.
• Reference — built-in tools that return the PushWard docs bundle and integration best practices, so the agent grounds its calls in the real API.

With a key from an organization, create_activity, update_activity, create_notification and create_scheduled_notification also take a target (groups, device tags or members) that narrows who gets the send, and update_activity takes clear_target to send an activity back to everyone.

Endpoint