One HTTPS POST to connect your website or App to AI workflows
No wiring up model SDKs, retries, or queuing yourself—your system sends a JSON, the fields map to workflow variables automatically, and once the AI is done it sends the result back to wherever you specify.
# One POST from your system triggers the workflow curl -X POST https://your-host/api/webhooks/{id}/trigger \ -H "X-API-Key: <your-api-key>" \ -H "Content-Type: application/json" \ -d '{ "user_feedback": "PDF export keeps failing after upgrading", "user_id": "u_84021", "app_version": "3.2.1" }' # Top-level fields map automatically to workflow variables of the same name # Then: AI classifies → drafts a reply → sends after human confirmation
Four ways to start a workflow
Clicking to run manually works too; the four below let workflows run automatically alongside your systems and schedule.
Results delivered where your team already works
When a workflow finishes, reports, alerts, and approval requests go to IM and email, or become file attachments.
For systems with no ready-made integration, connect an MCP Server to the Agent
An Agent in your workflow can attach any MCP Server as a tool and call it on its own during execution.
- Attach an MCP Server in an agent_based step, and the Agent calls it as a tool during execution
- Wrap an internal system in an MCP Server and Agents in your workflow can use it
- Complements sandboxed code steps: write code for deterministic calls, and leave open-ended tasks to an Agent with tools
What it looks like once connected
Two common patterns: both start with a single POST from your system, leaving key actions for human confirmation.
- A visitor submits the Contact Us form, and your backend POSTs the fields to the Webhook
- The AI reads the company name, message, and referring page to gauge lead intent
- High-intent leads stop at an approval step, a notification goes to Slack or email, and sales follows up after confirming
- A user submits feedback in the App, and the server forwards a POST
- A classifier step sorts feedback into bug, feature request, and billing issue
- The AI drafts a reply, sent to the user only after human confirmation
There's no catalog of hundreds of SaaS connectors here
Braidrun doesn't offer a drag-and-drop catalog of HTTP connectors. Calls to external systems go in code steps (7 languages, running in a Docker sandbox) or come in as MCP tools attached to an Agent.
Put another way: any system reachable over HTTP or an official SDK can go into a workflow—the cost is a few lines of code.