An AI agent API for creating and running scheduled agents from code

The REST API for crons and runs

The REST API covers the objects you see in the dashboard. You can create a cron from a plain English description, read the parsed schedule and steps, update it, pause it, start a run with input parameters and read the full run log, including every step input, output, retry and timing. Requests authenticate with API keys you create in workspace settings and can revoke at any time.

As an illustrative example, starting a run looks like a single request such as POST /api/v1/crons/{id}/runs with a small JSON body of inputs, and the response returns a run ID you can poll or receive later through an outgoing webhook. The exact paths and fields are listed in the API reference inside your workspace.

• Create crons from text so your product can offer scheduled agents to its own users
• Pass inputs per run, such as a company domain or a date range
• Read run logs to show results inside your internal tools
• Pause and resume crons from your deploy or incident tooling

Inbound webhooks with a shared secret

Not every job belongs on a clock. An inbound webhook gives a cron its own URL. When a form tool, a billing system or your own backend sends a request to that URL, the cron starts with the request body as input. Each webhook has a secret, and requests without the matching secret are rejected and logged.

On Teams you can also restrict inbound webhooks to an IP allowlist, so only your known servers can start runs. A cron can have both a schedule and an inbound webhook, for example a weekly full review plus an instant run whenever a new enterprise lead is created in HubSpot.

The MCP server for AI assistants and MCP scheduled tasks

Crons runs an MCP server, so AI assistants that support the Model Context Protocol can work with your crons directly. From the assistant you can list crons, read the latest results, create a new cron from a sentence and start a run. This is the simplest way to get MCP scheduled tasks: the assistant hands the job to Crons, and Crons keeps running it on schedule long after the chat window is closed.

1. 1Create an API key with the permissions the assistant should have.
2. 2Add the Crons MCP server to your assistant with that key.
3. 3Ask the assistant to list your crons or create one, for example "every morning summarize new support tickets tagged billing".
4. 4Review the created cron in the dashboard, including approval gates, before it runs for real.

Approval gates still apply to anything started through MCP. An assistant can draft a customer email inside a cron, but the email waits for a person to approve it. Learn more about the agent side on the scheduled AI agents page.

Code steps when you need exact logic

AI steps are great at reading messy pages and writing summaries. Some logic should be exact, such as currency conversion, a scoring formula or reshaping a JSON payload for your warehouse. Code steps run sandboxed JavaScript or Python inside the cron, with the previous step output as input and no access to other workspaces. Their output appears in the run log like any other step.

A common pattern is fetch, AI extraction, code step for calculation, memory diff, then delivery. The code is deterministic, so the same input gives the same result every run.

Security for programmatic access

• API keys are scoped to one workspace and can be revoked at any moment
• Webhook secrets are shown once and can be rotated without editing the cron
• IP allowlist for inbound webhooks on Teams
• SSO with SAML and an audit log of workspace changes on Teams
• Run logs record which key, webhook or assistant started each run

Read the full overview on the security page.