The Model Context Protocol (MCP) has made it easy to give AI agents tools. But when it comes to Apache DolphinScheduler, most public MCP servers stop at read-only operations — list projects, start a workflow, stop it. That's fine for a demo, not for real data-engineering work.
dolphin-mcp-pilot is a production-ready MCP server built for the operations reality.
Beyond read-only: 53+ tools
It exposes 53+ tools across projects, workflows, DAG creation, schedules, instances, resources, logs, monitoring and raw API passthrough:
- Create SQL and complex DAG workflows (multiple task types) in one call
- Manage schedules — create / online / offline / delete (cron-based)
- Full instance lifecycle control — pause / resume / rerun / rerun-from-failure / delete
- View task logs, force task success, skip failed task
- Resource content management, workflow version rollback and clone
- Raw API passthrough as a safety valve for uncovered edge cases
Built for real, multi-tenant operations
-
Two auth modes: API Token (
X-DS-Token) or User/Password (X-DS-User+X-DS-Password) - Multi-tenant HTTP mode: each caller can use its own credentials per request
- FastMCP, Python 3.10+, Apache-2.0 licensed
Quick start
git clone https://github.com/iflytek/dolphin-mcp-pilot.git
cd dolphin-mcp-pilot
cp .env.example .env # set DS_URL and DS_TOKEN (or DS_USER/DS_PASSWORD)
docker compose --profile dev up -d dolphin-mcp-pilot-dev
The service comes up at http://localhost:8001/mcp/ — point your MCP client (Claude Desktop, CodeBuddy, etc.) at it and your agent can start driving pipelines.
Why it matters
Data pipelines are exactly the kind of repetitive, error-prone, recovery-heavy work agents should help with — but only if they can do more than look. dolphin-mcp-pilot gives agents safe, credentialed hands on DolphinScheduler.
It's part of the iFLYTEK Astron open-source ecosystem. Repo: https://github.com/iflytek/dolphin-mcp-pilot
Top comments (0)