MCP — build automations with AI

Connect an AI assistant to your Cronium instance and create events and workflows from plain language — it drafts them, you approve, Cronium creates them.

Overview

Cronium ships a Model Context Protocol (MCP) server, so any MCP-capable AI app can build automations for you. You describe what you want; the assistant reads your Cronium's capabilities (tool types, your saved credentials, scheduling rules), drafts the events and workflow — filling in sensible defaults — and, once you approve, creates them.

For example, you can say:

“Create a workflow in Cronium: a SQL event that counts the users in the users database, then a Slack event that posts the count to #general at 8am every day.”

The assistant creates a two-step workflow — a SQL query feeding a scheduled Slack message (0 8 * * *) — with Local execution and a 30s timeout filled in. Everything lands as a draft you review and activate.

How it works

Discover

The assistant reads your tool credentials, the available actions, and Cronium's scheduling rules so it drafts valid config — not guesses.

Draft & validate

It can dry-run the plan (schemas, credentials, workflow graph) without changing anything, then show you the draft for approval.

Create as drafts

On your approval it creates the events and workflow — as drafts, so nothing runs until you review and activate them in Cronium.

There are two ways to connect, depending on which AI app you use.

Connect from the Claude web app

Add a custom connector in Claude pointing at your instance:

https://your-cronium-host/api/mcp

Cronium is a full OAuth 2.1 authorization server for this connector, so you just click Connect, sign in to Cronium, and approve — no token to paste. (Any tier of the Claude apps can add a custom connector; a paid plan is needed for more than one.) Alternatively, authenticate with a bearer API token as a custom header. Your instance must be reachable over HTTPS.

Connect from Claude Desktop or Claude Code

For local clients, run the bundled stdio server (apps/cronium-mcp in the Cronium repo) and point it at your instance with an API token. Claude Desktop — claude_desktop_config.json:

{
  "mcpServers": {
    "cronium": {
      "command": "node",
      "args": ["/path/to/apps/cronium-mcp/dist/index.js"],
      "env": {
        "CRONIUM_BASE_URL": "http://localhost:5001",
        "CRONIUM_API_TOKEN": "<your-token>"
      }
    }
  }
}

Claude Code:

claude mcp add cronium \
  --env CRONIUM_BASE_URL=http://localhost:5001 \
  --env CRONIUM_API_TOKEN=<your-token> \
  -- node /path/to/apps/cronium-mcp/dist/index.js

Tools the AI can use

get_capabilities — the enums, defaults, scheduling guidance, available tool actions, and your real credential/server ids.
validate_plan — dry-run a draft (events + optional workflow) without creating anything.
create_event / activate_event — create a single event (as a draft) and, for scheduled events, start it.
create_workflow — wire existing events into a workflow.
create_workflow_bundle — create several events and a workflow chaining them, in one step.

Authentication & scopes

The remote connector can use OAuth (recommended) or a bearer API token; the local server uses an API token. Create a token under Settings → API Tokens. When you create one, turn on “Limit to MCP” to make it least-privilege: such a token (and any OAuth connection) can only create and manage events and workflows — it can't touch the rest of your account, so a leaked connector token is contained.

Safety: drafts, approval, provenance

Everything is created as a draft. Nothing runs until you review it and activate it in Cronium.
Approval is in your hands. The assistant shows you the plan first; it only creates after you say so, and local clients also prompt before each action.
Provenance. Events and workflows created via MCP are tagged source = "mcp" and logged, so you can always tell what an AI agent created.
Scoping. Use an MCP-limited token/connection so the assistant can never act beyond creating automations.

Other MCP clients

MCP is an open standard, so the same server works with other MCP-capable clients (e.g. ChatGPT's custom connectors on Business/Enterprise plans, and developer tools like Cursor and VS Code). Remote clients use the /api/mcp endpoint; clients that support local servers can run the bundled stdio server.