ifly-workflow

This MCP server enables integration with iFlytek workflows by exposing the workflow execution capabilities through the MCP client. It supports a range of workflow features such as sequential, parallel, looped, and nested execution, as well as multiple node types and a hook mechanism for streaming output. By registering the server under an MCP client, you can publish and manage workflows, then trigger them via the MCP protocol and receive results as part of your chat or automation flows. To use it, prepare your MCP config and point the server to a YAML configuration that specifies workflow metadata and authentication information, then interact with your workflows through the MCP interface and the bound applications in your environment.

Prerequisites:

• Python 3.8+ installed on your system
• Git available to fetch packages
• Access to the internet to pull the MCP server and dependencies
• Optional: pipx for isolated Python CLI installations

Installation steps:

1. Install the MCP client (uvx) or ensure it is available in your environment:

   • With pipx (recommended):

     pipx install uvx
     

   • Or using Python's pip:

     python -m pip install uvx
     

2. Install or run the MCP server from its GitHub source using uvx:

   • From GitHub (recommended):

     uvx --from git+https://github.com/iflytek/ifly-workflow-mcp-server ifly_workflow_mcp_server
     

3. Prepare a config.yaml for your workflows and place it where the server can read it. For MCP integration, you typically set an environment variable to point to this config:

   - flow_id: 'flow id'              # required
     name: 'flow name'               # optional
     description: 'flow description' # optional
     api_key: 'API Key:API Secret'   # required
   

4. If you are using the sample in the README as a guide, you can bind CONFIG_PATH to your local config file:

   export CONFIG_PATH=/path/to/your/config.yaml
   

   Then run the MCP registration snippet as shown in the example to expose the server under ifly-workflow-mcp-server.

5. Validate installation by listing/triggering a simple workflow through your MCP client and verify the responses are streamed or returned as expected.

Tips and notes:

• The server relies on a CONFIG_PATH pointing to a YAML config with authentication details; ensure API keys or credentials are kept secure and not committed to source control.
• If you encounter network or dependency errors, verify Python environment compatibility and that uvx is correctly installed in your PATH or via your chosen enclosure (e.g., pipx).
• When debugging, you can temporarily set CONFIG_PATH to a test workflow configuration to validate MCP interactions before going live.
• The MCP server supports streaming outputs via the hook mechanism; ensure your client can handle incremental updates if you rely on real-time results.
• If you need to switch from uvx to a node-based or other runtime, adjust the mcp_config accordingly (not applicable for this Python/uvx-based server).