AI agents and MCP

HelloCron ships one package for AI assistants and coding agents: three Agent Skills that teach the model how to use the shell client, design monitoring and call the API, plus a remote MCP server that lets the model manage monitors on your account. Everything lives in the public hellocron/agent-plugin repository.

What you get

Component	What it does	Works in
Skills (`SKILL.md`)	Instructions the agent loads on demand: install and use `hellocron.sh`, pick intervals and grace, wrap cron jobs, write bundles, call the API	Claude Code, Codex CLI, Cursor, GitHub Copilot, Gemini CLI and every tool that reads the Agent Skills format
MCP server	Tools such as `list_monitors`, `create_monitor`, `apply_monitor_bundle`, `ping`, backed by your API keys	Claude (Desktop, Code, claude.ai), ChatGPT (Developer Mode), Codex, Cursor, Gemini CLI, VS Code
Plugin package	Skills and the MCP connection bundled in the Agent Plugins format so one install adds both	Claude Code, Codex, ChatGPT
`llms.txt`	Index of these docs for models: `/llms.txt` and the full text at `/llms-full.txt`	any agent with web access

The MCP server never stores your keys. Each request is forwarded to the public API with the Authorization header you configured in the client, so plan limits, rate limits and key scopes apply exactly as they do for curl.

Keys

Create keys in the panel under Settings → API keys:

an ingest key (ck_...) for pings; safe to put on monitored servers,
a management key (mk_...) for listing, creating, updating, deleting and applying monitors; keep it on your workstation or in CI secrets only.

Give the MCP client the management key when you want the agent to manage monitors, the ingest key when it should only send pings.

Claude Code

claude plugin marketplace add hellocron/agent-plugin
claude plugin install hellocron@hellocron

The plugin registers the skills (/hellocron:hellocron-cli, /hellocron:hellocron-monitoring, /hellocron:hellocron-api) and the MCP server. Set the key once:

claude mcp add --transport http hellocron https://mcp.hellocron.com/mcp \
  --header "Authorization: Bearer mk_..."

Codex CLI and ChatGPT

codex plugin marketplace add hellocron/agent-plugin

Then enable the HelloCron plugin in Codex or ChatGPT. When asked for the connection, paste https://mcp.hellocron.com/mcp and your key. Without the plugin, ChatGPT Developer Mode accepts the same URL as a custom MCP connection, and a Custom GPT can use the API directly through Actions: import https://docs.hellocron.com/openapi.yaml and choose API key authentication with the Authorization: Bearer header.

Cursor, Gemini CLI, VS Code, other MCP clients

Add the server to the client’s MCP configuration:

{
  "mcpServers": {
    "hellocron": {
      "type": "streamable-http",
      "url": "https://mcp.hellocron.com/mcp",
      "headers": { "Authorization": "Bearer mk_..." }
    }
  }
}

For the skills, copy the skills/ directory of the repository into the location your tool reads (.cursor/skills/, .gemini/skills/, .agents/skills/ or the project’s skills/; see the tool’s documentation). The files are plain Markdown and need no build step.

Example prompts

“Install the HelloCron client on this server, verify it with doctor and wrap every job in /etc/cron.d so failures alert me.”
“Create an HTTP check for https://api.example.com/health every minute expecting a 200 and the text ok, and an SSL check for example.com.”
“Export my monitors as a bundle, add a nightly db-backup ping monitor with one hour of grace and apply it as a dry run first.”

Tools exposed by the MCP server

Tool	Endpoint	Notes
`get_account`	`GET /api/v1/account`	read-only
`list_monitors`, `get_monitor`	`GET /api/v1/monitors[/{uuid}]`	read-only
`export_monitors`	`GET /api/v1/monitors/export`	read-only
`create_monitor`, `update_monitor`	`POST`, `PUT /api/v1/monitors`	manifests, see Monitors
`apply_monitor_bundle`	`POST /api/v1/monitors/apply`	`dry_run` supported, never deletes
`delete_monitor`	`DELETE /api/v1/monitors/{uuid}?confirm=true`	destructive, requires `confirm: true`
`ping`	`POST /ping`	needs an ingest key
`get_health`	`GET /health`	reachability

Tools carry MCP annotations (readOnlyHint, destructiveHint) so clients can ask for confirmation before writes. The full contract is the OpenAPI document.