Workflows (Growth+)

Reusable recipes the agent — or you — can run on demand to automate multi-step actions.

A workflow is a named list of steps that Coffield.io runs for you in the background. Each run is logged so you can see exactly what happened.

#What can a workflow do?

In v1, three step types are supported. Every one is safe by design — none of them can run shell commands, touch arbitrary URLs, or leave your tenant.

More step types (notify a teammate, create a Kanban card, push to a CRM) are coming in future versions.

#Plan availability

If you try to run a workflow above your cap, you'll see a friendly "Daily workflow cap reached" message and the run is not started — no surprises on your bill, no half-finished side effects.

#Finding the page

Sign in to your dashboard, then in the left menu under AI Agents you'll see two new items:

• Workflows — the list of recipes available to your tenant. Each row has a Run now button.
• Workflow Runs — every run ever started, with status (queued / running / succeeded / failed), step counts, and who triggered it.

#Running a workflow manually

1. Open AI Agents → Workflows.
2. Find the recipe you want to run.
3. Click Run now.
4. Confirm the dialog.
5. A toast appears: "Workflow run #N queued."
6. Open AI Agents → Workflows Runs to watch its status update.

A run that finishes with status succeeded ran every step without error. A run that ends failed stopped on the first error (unless the recipe was authored to continue past errors).

#Triggering a workflow from the REST API (Pro)

If you're on the Pro plan you can also trigger workflows programmatically. See the REST API chapter for full details and a list of endpoints. The short version:

curl -X POST \
  -H "Authorization: Bearer <your-PAT>" \
  -H "Content-Type: application/json" \
  -d '{"input":{"campaign":"Spring Promo"}}' \
  https://your-domain.com/api/v1/workflows/123/runs

A successful trigger returns HTTP 202 Accepted with the new run's ID. The run is processed in the background — poll GET /api/v1/workflow-runs/{id} to check status.

#Triggering a workflow from an MCP client (Claude Desktop, n8n, Zapier, Make)

If you've connected an MCP client (see External Integrations & MCP tokens), every issued token that includes the Workflow server can also trigger workflows. The MCP server exposes these tools:

Tier and rate-limit gates are identical to the REST API: triggering requires Growth or above and counts against your tenant's daily cap. Cross-tenant workflow_id or run_id values are silently rejected with a not-found error — your token can only see its own tenant's data.

#Pairing with webhooks

The most powerful step is fire_webhook_event. A typical pattern:

1. Register a webhook endpoint under AI Agents → Webhook Endpoints subscribed to a custom event like campaign.kickoff.
2. Author a workflow recipe that fires campaign.kickoff with the payload your downstream tool expects.
3. Trigger the workflow manually or from the REST API.

Every webhook delivery is signed (HMAC-SHA256) and logged under AI Agents → Webhook Deliveries so you can verify exactly what was sent.

#How a recipe is authored

In v1, recipes are stored as JSON inside the workflow row. The dashboard shows the recipe but does not yet include a visual editor. Customers on Growth and Pro plans should ask their account manager to author recipes for them; future releases will add a drag-and-drop editor.

A recipe looks like this:

{
  "steps": [
    { "type": "log", "message": "Kicking off Spring Promo" },
    { "type": "delay", "seconds": 2 },
    {
      "type": "fire_webhook_event",
      "event": "campaign.kickoff",
      "payload": { "campaign": "spring-2026", "owner": "marketing" }
    }
  ]
}

Recipes are capped at 50 steps to keep runs short and predictable.

#Continuing past errors

By default a workflow stops on the first failed step. Add "continue_on_error": true at the root of the recipe to keep going past failures:

{
  "continue_on_error": true,
  "steps": [ ... ]
}

This is useful when each step is independent (for example, firing the same event to several different downstream tools).

#Safety & limits

• Recipes can only contain the three step types listed above. Anything else is rejected with step_type_not_allowed.
• The total step count per recipe is capped at 50.
• Daily run counts are enforced per tenant — see plan table above.
• Every run row is permanent and tied to your tenant; no run can read or write another tenant's data.
• Webhook deliveries triggered by fire_webhook_event go through the same SSRF/private-network protection as ordinary webhook deliveries — only HTTPS endpoints on public hostnames are allowed.

#Troubleshooting

If you need a workflow authored, a recipe extended, or a higher daily cap, email support@coffield.io with your tenant name and what you're trying to automate.