Multi-agent workflows, composed visually

Build DAG-based orchestrations that chain tools, invoke agents, branch on conditions, fan out in parallel, and loop until done. Connect your Fold Agent alongside any external A2A or MCP service.

Nine building blocks

Drag nodes onto the canvas and draw edges between them. Each node type handles a different step in your workflow.

Tools

Execute any deployed MCP tool. Same dispatch path as direct invocations, with input guardrails applied automatically.

Agents

Invoke your workspace's Fold Agent with a prompt template, or call an external A2A agent by selecting a registered skill.

Conditions

Branch the workflow based on the output of a previous node. Route to different paths depending on values, scores, or categories.

Parallel

Fan out to multiple nodes concurrently. Merge results when all branches complete or take the first response.

Loops

Repeat a set of steps until a condition is met or a max iteration count is reached. Useful for iterative refinement.

Sub-orchestrations

Nest one orchestration inside another. Compose complex workflows from reusable building blocks without duplication.

Connect any agent

Register external A2A agents or MCP servers. The platform fetches their capabilities automatically and makes their skills available in the orchestration builder.

A2A discovery

Paste an agent URL and fold.run fetches its agent card. Skills, input modes, and capabilities are cached and refreshable on demand.

MCP tool listing

Connect to any MCP endpoint. The platform sends initialize and tools/list to discover available tools, then makes them selectable in agent nodes.

Trigger from anywhere

Start orchestrations manually, on a schedule, via webhook, or through A2A and MCP calls.

Webhook and schedule

Fire on incoming HTTP requests or cron expressions. Webhook triggers get a dedicated URL at your organization's subdomain.

MCP and A2A

Active orchestrations auto-register as MCP tools and A2A skills. Any compatible agent can discover and invoke them without extra configuration.

Built for reliability

Orchestrations execute with per-node result tracking and optional durable execution for crash recovery.

$ curl -X POST https://api.fold.run/orchestrations/orch_abc/execute \
-H "Authorization: Bearer $TOKEN" \
-d '{ "input": { "message": "Help with billing" } }'
 
{
"status": "completed",
"node_results": [
{ "node_id": "trigger_1", "status": "success", "duration_ms": 0 },
{ "node_id": "agent_1", "status": "success", "duration_ms": 450 },
{ "node_id": "output_1", "status": "success", "duration_ms": 1 }
],
"duration_ms": 451
}

Run history

Every execution is recorded with per-node input, output, and timing. Inspect individual runs from the Console or the API.

Export and import

Export orchestrations as portable JSON. Import into other workspaces or share with teammates. Imported orchestrations start in draft status.

← DeployProtocols →

Build your first orchestration

Free tier. No credit card. Compose tools and agents into workflows in minutes.

Get Started