An MCP server for storing, querying, and creating YAML workflow playbooks for LLM agents from a local workflows library. It supports listing permanent workflows, retrieving workflows with global instructions, and creating permanent or temporary workflows over STDIO or Streamable HTTP.
From the registry: Store, query, and create YAML workflow playbooks for LLM agents.
Please install the `workflows-mcp-server` MCP server into my current AI client (that's you).
Optional prerequisites:
- Install Node.js or Bun — Install Bun v1.3.2+ or Node.js v24+ so you can run the server. Prefer the npx path for portability, which requires Node.js and npm/npx. (https://bun.sh/)
- Prepare a local workflows directory — Create or choose a local directory containing YAML workflow files, or use the repository's bundled workflows-yaml seed content. The server reads from a local workflows directory.
Canonical MCP server config (stdio transport):
- command: `npx`
- args: ["-y","@cyanheads/workflows-mcp-server@latest"]
- required environment variables:
- `MCP_TRANSPORT_TYPE`: Transport to use for the MCP server. Set to stdio for client config usage. (example: `stdio`)
- `WORKFLOWS_DIR`: Absolute or relative path to the workflows root directory containing your YAML workflow library. Although the README documents a default, in practical MCP client setup you should set this explicitly to the directory you want the server to read. (example: `<your-workflows-dir>`)
- optional environment variables:
- `GLOBAL_INSTRUCTIONS_PATH`: Path to the global instructions markdown file. If omitted, it is derived from WORKFLOWS_DIR as <WORKFLOWS_DIR>/global_instructions.md. (example: `<your-global-instructions-path>`)
- `WATCHER_DEBOUNCE_MS`: Milliseconds to debounce filesystem change events before rebuilding the workflow index. (example: `<your-watcher-debounce-ms>`)
- `MCP_HTTP_PORT`: Port for HTTP mode. Not needed for stdio transport. (example: `<your-http-port>`)
- `MCP_AUTH_MODE`: Authentication mode: none, jwt, or oauth. (example: `<your-auth-mode>`)
- `MCP_LOG_LEVEL`: Log level in RFC 5424 format. (example: `<your-log-level>`)
- `OTEL_ENABLED`: Enable OpenTelemetry instrumentation. (example: `<true-or-false>`)
Note: The README also supports Streamable HTTP. To run in HTTP mode, set MCP_TRANSPORT_TYPE=http and the server listens at <your-server-url>/mcp, with the example path shown as http://localhost:3010/mcp. However, no canonical MCP client HTTP config snippet with headers was provided, so stdio is the safest extracted config.
Add this MCP server to my current client's config in the correct format for you. If you need secrets or credentials I haven't provided, ASK me — do not invent values or leave raw placeholders. After adding it, tell me how to verify the server is connected.WORKFLOWS_DIRPath to the workflows root directory used as the local workflow library.GLOBAL_INSTRUCTIONS_PATHPath to the global instructions markdown file prepended to workflow_get responses.WATCHER_DEBOUNCE_MSDebounce interval for filesystem change events before rebuilding the index.MCP_TRANSPORT_TYPETransport type for the server, either stdio or http.MCP_HTTP_PORTPort used when running the HTTP server.MCP_AUTH_MODEAuthentication mode for the server.MCP_LOG_LEVELLogging level for the server.OTEL_ENABLEDEnables OpenTelemetry instrumentation.Real-time infrastructure monitoring with metrics, logs, alerts, and ML-based anomaly detection.
Trending hip-hop artist momentum scores across four cultural dimensions.
Web scraping with stealth HTTP, real browsers, and Cloudflare bypass. CSS selectors supported.