Ifly_workflow

You can expose and call iFlytek workflows through an MCP server to integrate AI-driven processes with external data and tools. This server enables scheduling and executing workflows with multiple node types, real-time streaming, and flexible execution modes, making it easier to build automated, end-to-end AI workflows.

How to use

You interact with the iFlytek workflow MCP server by connecting an MCP client to run and manage workflows. Start a workflow, publish it as an API, and bind it to an application to obtain authentication details. You can trigger sequential, parallel, looping, or nested executions, and you can leverage multi-model choices at key workflow stages. Real-time streaming output is supported through a hook mechanism, so you can monitor progress as the workflow runs.

How to install

Prerequisites: ensure you have a runtime capable of running MCP servers and a client that can interact with MCP endpoints. You will configure a local or remote MCP client to load and run the iFlytek workflow server.

1. Create a persistent client configuration file for MCP. You will specify the MCP server configuration under the mcpServers section. Use the provided JSON structure as a starting point.

How to install

{
  "mcpServers": {
    "ifly_workflow_mcp": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/iflytek/ifly-workflow-mcp-server",
        "ifly_workflow_mcp_server"
      ],
      "env": {
        "CONFIG_PATH": "$CONFIG_PATH"
      }
    }
  }
}

How to install

If you need a concrete example for your local client, you can use an explicit path for the CONFIG_PATH in your environment and place the JSON above into your client’s config file (for example claude_desktop_config.json or mcp.json).

How to install

{
  "mcpServers": {
    "ifly-workflow-mcp-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/iflytek/ifly-workflow-mcp-server",
        "ifly_workflow_mcp_server"
      ],
      "env": {
        "CONFIG_PATH": "/Users/yourname/Projects/config.yaml"
      }
    }
  }
}

How to install

If you are following a multi-step setup, ensure you finalize the start configuration so the MCP client can launch the server with the specified CONFIG_PATH. The example config shows how to point the client to a local or cloud-based workflow server and pass the required environment variable for the configuration path.

Additional sections

Notes and tips: The MCP server supports 14 types of workflow nodes and various execution modes, including sequential, parallel, loop, and nested execution. You can enable streaming output via the Hook Mechanism for real-time progress updates. To expand capabilities, you can rely on multi-model support within a hybrid architecture, allowing different models to be used at critical workflow steps.

Additional sections

To publish and authenticate workflows: create a bot, publish the workflow, test it in conversation, then choose Publish as API and configure the binding to an application. Retrieve the workflow ID and authentication details for subsequent API calls.