Display Name
Pushary
Category
Tooling
Sub-Category
General
Primary Link
https://github.com/Pushary/pushary-skill
Author Name
Pushary
Author Link
https://github.com/Pushary
License
MIT
Other License
No response
Description
Pushary is an MCP server and permission-hook integration for Claude Code that sends a push notification when a session finishes or needs a decision, and lets the user approve, deny, or answer from a phone. It applies per-tool permission policies (auto-approve, deny, escalate, timeout) and keeps an exportable audit trail of approval decisions across multiple agents. The skill and CLI hooks are MIT-licensed; notification delivery and approvals are relayed through Pushary's hosted API, which requires a Pushary account and API key.
Validate Claims
Install with npx skills add Pushary/pushary-skill, create a Pushary account to get an API key, and add the Pushary MCP server to ~/.claude/settings.json (the exact JSON block is in the repo README). For the permission hook, run npx @pushary/agent-hooks to register a PreToolUse hook. Uninstall by removing the MCP server entry and the hook from settings.json and deleting the npm package. No build script is required.
Specific Task(s)
Give Claude Code a task that exercises both paths: one long-running step that finishes after a few minutes (a completion alert), and one destructive shell command gated by a permission policy set to escalate (an approval request to the phone).
Specific Prompt(s)
Run the full test suite, then delete the dist directory and force-push to main. With a policy that escalates destructive git and filesystem commands, Claude pauses and Pushary pushes an approve or deny prompt to your phone; when the run completes it pushes a done notification.
Additional Comments
Network requests: the skill and hook send notification payloads and approval requests to Pushary's hosted API (https://pushary.com/api/mcp), not only to the Anthropic API. There is no telemetry beyond delivering the notifications the agent triggers. It does not require --dangerously-skip-permissions; the PreToolUse hook returns allow / deny / ask and is built to gate permissions, not bypass them. The agent-side skill, the @pushary/agent-hooks CLI (MIT), and the Hermes plugin are open source; the backend that stores policies and relays push delivery is a hosted commercial service that requires an account.
Recommendation Checklist
Display Name
Pushary
Category
Tooling
Sub-Category
General
Primary Link
https://github.com/Pushary/pushary-skill
Author Name
Pushary
Author Link
https://github.com/Pushary
License
MIT
Other License
No response
Description
Pushary is an MCP server and permission-hook integration for Claude Code that sends a push notification when a session finishes or needs a decision, and lets the user approve, deny, or answer from a phone. It applies per-tool permission policies (auto-approve, deny, escalate, timeout) and keeps an exportable audit trail of approval decisions across multiple agents. The skill and CLI hooks are MIT-licensed; notification delivery and approvals are relayed through Pushary's hosted API, which requires a Pushary account and API key.
Validate Claims
Install with
npx skills add Pushary/pushary-skill, create a Pushary account to get an API key, and add the Pushary MCP server to ~/.claude/settings.json (the exact JSON block is in the repo README). For the permission hook, runnpx @pushary/agent-hooksto register a PreToolUse hook. Uninstall by removing the MCP server entry and the hook from settings.json and deleting the npm package. No build script is required.Specific Task(s)
Give Claude Code a task that exercises both paths: one long-running step that finishes after a few minutes (a completion alert), and one destructive shell command gated by a permission policy set to escalate (an approval request to the phone).
Specific Prompt(s)
Run the full test suite, then delete the dist directory and force-push to main. With a policy that escalates destructive git and filesystem commands, Claude pauses and Pushary pushes an approve or deny prompt to your phone; when the run completes it pushes a done notification.
Additional Comments
Network requests: the skill and hook send notification payloads and approval requests to Pushary's hosted API (https://pushary.com/api/mcp), not only to the Anthropic API. There is no telemetry beyond delivering the notifications the agent triggers. It does not require --dangerously-skip-permissions; the PreToolUse hook returns allow / deny / ask and is built to gate permissions, not bypass them. The agent-side skill, the @pushary/agent-hooks CLI (MIT), and the Hermes plugin are open source; the backend that stores policies and relays push delivery is a hosted commercial service that requires an account.
Recommendation Checklist