Create an agent from authoring code

Tool: create_workflow. Requires Editor, Admin, or Owner in the applicable workspace and a connected OAuth session. Accepts the complete supported TypeScript authoring source in code. Start from code returned by get_workflow, change its workflow name and desired behavior, and submit the full source. The name in that source is required. This tool parses the source; it is not a general TypeScript execution service. For a fresh agent without code, use create_workflow_from_template or create_single_prompt_agent. Success has created: true, lifecycle draft, and version 1 published for the stored snapshot. Failures include parse_error, validation_error, graph_validation, missing_name, and trigger_path_conflict. Correct the full source before retrying. The tool does not call a model/provider or publish the agent; URL checks may resolve hostnames. The example below is complete authoring text for a two-node agent. Send it as the code string; Agent Builder parses Workflow, startCall, and endCall directly. You do not need to install an SDK or execute this text locally.

Arguments

Name	Type	Required	Defaults and limits
`code`	string	Yes	See full schema

Example arguments

Replace sample resource IDs with IDs returned in your selected workspace.

{
  "code": "const wf = new Workflow({ name: \"MCP workflow assistant\" });\nconst welcome = wf.addTyped(startCall({ name: \"Welcome\", prompt: \"Ask how you can help.\" }));\nconst done = wf.addTyped(endCall({ name: \"Done\", prompt: \"Thank the caller.\" }));\nwf.edge(welcome, done, { label: \"done\", condition: \"The caller is finished.\" });"
}

Result and errors

The schema below describes MCP structuredContent, including the result wrapper where present. Your client may also display a text representation. Read success flags and structured errors before continuing; transport success alone does not prove the operation succeeded. Authentication, missing membership, and insufficient role can fail before a tool result is produced. See troubleshooting. Effect: Changes state. Retry: Do not blindly repeat successful calls. Discovery annotations describe behavior; they do not grant permission or replace confirmation for consequential actions.

Input schema

{
  "additionalProperties": false,
  "properties": {
    "code": {
      "type": "string"
    }
  },
  "required": [
    "code"
  ],
  "type": "object"
}

Result schema

{
  "properties": {
    "result": {
      "anyOf": [
        {
          "properties": {
            "assistant_mode": {
              "enum": [
                "single_prompt",
                "workflow"
              ],
              "type": "string"
            },
            "created": {
              "const": true,
              "type": "boolean"
            },
            "edge_count": {
              "type": "integer"
            },
            "lifecycle_status": {
              "enum": [
                "draft",
                "active",
                "paused",
                "archived"
              ],
              "type": "string"
            },
            "name": {
              "type": "string"
            },
            "node_count": {
              "type": "integer"
            },
            "version_number": {
              "type": "integer"
            },
            "version_status": {
              "enum": [
                "draft",
                "published"
              ],
              "type": "string"
            },
            "workflow_id": {
              "type": "integer"
            }
          },
          "required": [
            "created",
            "workflow_id",
            "name",
            "lifecycle_status",
            "version_status",
            "version_number",
            "assistant_mode"
          ],
          "type": "object"
        },
        {
          "properties": {
            "created": {
              "const": false,
              "type": "boolean"
            },
            "error": {
              "type": "string"
            },
            "error_code": {
              "type": "string"
            },
            "errors": {
              "items": {
                "additionalProperties": true,
                "type": "object"
              },
              "type": "array"
            },
            "trigger_paths": {
              "items": {
                "type": "string"
              },
              "type": "array"
            }
          },
          "required": [
            "created",
            "error_code",
            "error"
          ],
          "type": "object"
        }
      ]
    }
  },
  "required": [
    "result"
  ],
  "type": "object"
}

Get started

Use MCP

Workspaces

Agents

Calls and outcomes

Phone numbers

Resources

Authoring guidance

Documentation

Arguments

Example arguments

Result and errors

​Arguments

​Example arguments

​Result and errors

Arguments

Example arguments

Result and errors