workflows

The MCP Workflow Orchestration Server is a specialized MCP server that lets your AI agents discover, inspect, and execute complex multi-step workflows defined in YAML. It exposes a set of workflow tools that can list available workflows, fetch full workflow definitions, create new workflows, or generate temporary, non-listed workflows for planning and task decomposition. This enables dynamic, reusable orchestration of tasks without embedding rigid logic in the agent.

To use it, connect your MCP client to the workflows-mcp-server and leverage the provided tools by name. For example, use workflow_return_list to discover workflows by category or tag, workflow_get_instructions to retrieve a specific workflow’s full YAML definition (with versioning and global instructions injected), workflow_create_new to add a new workflow to the library, and workflow_create_temporary to draft a temporary plan that can be referenced by name during a session. The server handles indexing, validation, and secure execution of the workflows, making it easier to build robust automation pipelines for your agents.

Prerequisites:

• Node.js (>= 20.0.0)
• npm (comes with Node.js)

Option A: Use via MCP client (recommended for production)

1. Ensure you have a working MCP client configuration file (e.g., cline_mcp_settings.json).
2. Add the server entry (already included in the repository example) to your MCP client settings:

{
  "mcpServers": {
    "workflows-mcp-server": {
      "command": "npx",
      "args": ["workflows-mcp-server"],
      "env": {
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Option B: Run directly for development

1. Clone the repository and install dependencies:

git clone https://github.com/cyanheads/workflows-mcp-server.git
cd workflows-mcp-server
npm install

2. Build or run the server in development mode (as defined by the template you’re using). If there is a build step, run:

npm run build

3. Start the server (depending on the package.json scripts, usually something like):

npm run start

Tip: For development, you can install dependencies locally and run the server directly via npm, which will also help you test the tooling API endpoints and YAML workflow parsing.

Notes and tips:

• The server relies on YAML-defined workflows. Ensure your workflow files are properly formatted and indexed on startup.
• Global instructions can be injected into every workflow to maintain consistent execution strategies.
• Temporary workflows are useful for planning or multi-step tasks that don’t require permanent publication. They can be called by name during a session.
• Environment variable MCP_LOG_LEVEL controls verbosity of logs; set to debug for development and info/warn/error for production.
• When running via npx, the first run will install the package if not already present in the environment.
• If you extend or modify workflows, consider re-indexing or restarting the server to ensure the new files are picked up.