Repository Name
dolphin-mcp-pilot
Repository Description
A production-ready MCP server for Apache DolphinScheduler. Exposes 58 tools for projects, workflows, DAG creation, schedules, instances, resources, logs and monitoring, so AI agents can operate DolphinScheduler through natural language. It acts as an AI-agent integration layer over the DolphinScheduler REST API, not a replacement for it.
Visibility
Public
Project Admins
@Hui-of-limin
@charleswillicks
Additional Context
Repository Information
- Name: dolphin-mcp-pilot
- Visibility: Public
- Description: Operate Apache DolphinScheduler from AI agents via MCP
Project Details
dolphin-mcp-pilot is an MCP server that exposes Apache DolphinScheduler operations
as 58 discrete tools. AI agents (Claude Desktop, CodeBuddy, Cursor and any other
MCP client) can create workflows, configure schedules, run backfills, inspect task
instances and pull failure logs without touching the REST API directly.
It does not replace the DolphinScheduler API or the scheduler core. Instead, it works
as an AI-agent integration layer: translating agent intents into validated
DolphinScheduler REST calls and returning structured, agent-friendly results.
Example:
Input (agent intent):
Backfill the daily_sales workflow for 2026-01-01 through 2026-01-05, one day at a time.
Tool call emitted:
ds_complement_data(
project_code=...,
workflow_code=...,
complement_start_date="2026-01-01",
complement_end_date="2026-01-05",
run_mode="RUN_MODE_SERIAL"
)
Result: five instances generated in strict ascending date order.
Features:
- 58 tools across 10 categories: connectivity, projects, datasources, workflows,
advanced workflow ops, schedules, instances, resources, monitoring, raw API passthrough
ds_help for in-session tool discoverynext_action hints: when an instance is RUNNING or FAILURE, the response tells the
agent to drill into task instances instead of polling the workflow-level state- Deterministic serial backfill: uses a date-range payload so DolphinScheduler emits
instances day by day in order
- Dual auth: username/password or session token, injected per request via headers
- Three transports: stdio, SSE, streamable HTTP
Why this belongs in the iflytek org
Apache DolphinScheduler is one of the most widely adopted open source data
orchestration systems (Apache top-level project, 12k+ GitHub stars, used across
finance, internet and enterprise data platforms). iflytek is a heavy user of it.
There is currently no mature AI-agent integration for DolphinScheduler. The closest
existing projects are ocean-zhc/dolphinscheduler-mcp and lukaa077/ds-mcp; both
expose a much smaller tool surface and lack the operational fixes we needed in
production. Open sourcing this contributes our AI-plus-orchestration practice back
to the DolphinScheduler community.
iflytek Landscape
- Board: Model & Agent System
- Track: Toolchain (3.2.3)
Technical Profile
| Item |
Value |
| Language |
Python 3.10+ |
| Lines of code |
5,209 |
| Tools |
58 |
| Dependencies |
5 (mcp, anyio, uvicorn, starlette, pydantic) |
| Tests |
19 unit tests, 29 smoke checks |
| CI |
unit tests, ruff, bandit, secret scan, build |
License
Apache License 2.0
- ✅ Copyright headers applied to all 21 Python source files
- ✅ All dependencies are MIT or BSD-3-Clause; no copyleft, no license conflict
- ✅ 100% original code, no third-party code copied in
Compliance Self-Check
- ✅ No confidential platform logic, model weights, training algorithms or partner-specific code
- ✅ No internal hostnames, credentials or tokens; verified by a secret-scan CI step
- ✅ Configuration is placeholder-only (
your-ds-host, your-ds-user)
- ✅ Intellectual property held by iFLYTEK CO., LTD.
Acknowledgements
- ✅ I have reviewed the community guidelines
- ✅ I will be the primary maintainer or know who will be
Repository Name
dolphin-mcp-pilot
Repository Description
A production-ready MCP server for Apache DolphinScheduler. Exposes 58 tools for projects, workflows, DAG creation, schedules, instances, resources, logs and monitoring, so AI agents can operate DolphinScheduler through natural language. It acts as an AI-agent integration layer over the DolphinScheduler REST API, not a replacement for it.
Visibility
Public
Project Admins
@Hui-of-limin
@charleswillicks
Additional Context
Repository Information
Project Details
dolphin-mcp-pilot is an MCP server that exposes Apache DolphinScheduler operations
as 58 discrete tools. AI agents (Claude Desktop, CodeBuddy, Cursor and any other
MCP client) can create workflows, configure schedules, run backfills, inspect task
instances and pull failure logs without touching the REST API directly.
It does not replace the DolphinScheduler API or the scheduler core. Instead, it works
as an AI-agent integration layer: translating agent intents into validated
DolphinScheduler REST calls and returning structured, agent-friendly results.
Example:
Input (agent intent):
Tool call emitted:
Result: five instances generated in strict ascending date order.
Features:
advanced workflow ops, schedules, instances, resources, monitoring, raw API passthrough
ds_helpfor in-session tool discoverynext_actionhints: when an instance is RUNNING or FAILURE, the response tells theagent to drill into task instances instead of polling the workflow-level state
instances day by day in order
Why this belongs in the iflytek org
Apache DolphinScheduler is one of the most widely adopted open source data
orchestration systems (Apache top-level project, 12k+ GitHub stars, used across
finance, internet and enterprise data platforms). iflytek is a heavy user of it.
There is currently no mature AI-agent integration for DolphinScheduler. The closest
existing projects are
ocean-zhc/dolphinscheduler-mcpandlukaa077/ds-mcp; bothexpose a much smaller tool surface and lack the operational fixes we needed in
production. Open sourcing this contributes our AI-plus-orchestration practice back
to the DolphinScheduler community.
iflytek Landscape
Technical Profile
License
Apache License 2.0
Compliance Self-Check
your-ds-host,your-ds-user)Acknowledgements