mirror of https://github.com/apache/superset.git synced 2026-05-12 11:25:56 +00:00

Files

Amin Ghadersohi 364af98c04 update: create_chart: Remove x_axis from groupby if present, update docs and tests

- Updates create_chart logic to automatically remove x_axis from groupby for ECharts timeseries charts, preventing duplicate dimension usage.
- Updates and expands unit test to verify x_axis is excluded from groupby, using improved test mocks for accurate backend simulation.
- Updates documentation (README.md, README_ARCHITECTURE.md, README_PHASE1_STATUS.md, README_SCHEMAS.md) to clarify create_chart tool behavior and schema, including new groupby/x_axis handling.
- No breaking changes to tool signatures; behavior is now more robust and LLM-friendly.

2025-07-30 14:20:39 -04:00

4.9 KiB

Raw Blame History

Superset MCP Service – Phase 1 Status Update

Background

The Model Context Protocol (MCP) is a new protocol for exposing high-level, structured actions in Superset, designed for LLM agents and automation. Phase 1 delivers a foundational, extensible MCP service in Superset, leveraging internal APIs (DAOs/commands) and providing a versioned, developer-friendly interface for both Apache and Preset use cases. (SIP-171)

Phase 1 Objectives (from SoW/SIP-171)

Standalone MCP service, config flag, CLI, modular, stateless
Strong typing: all actions use DAOs/commands and Pydantic schemas
Clear extension points for Preset-specific auth, RBAC, and logging
3+ high-value MCP actions (list, info, mutation)
Developer experience: easy to run, clear docs, tests
Auth/RBAC/logging hooks stubbed, ready for enterprise
Out of scope: full RBAC, impersonation, logging, external identity provider integration

What’s Delivered (Phase 1)

Service infrastructure: ASGI-based FastMCP server, config flag, CLI (superset mcp run), stateless
Strong typing: All tool input/output uses Pydantic v2 models with field descriptions
Modular tools: Grouped by domain (dashboard/, dataset/, chart/, system/), shared abstractions in model_tools.py and middleware.py
Abstracted, reusable tools: List, info, and available filters tools are now abstracted and reusable across datasets, charts, and dashboards
Tool registration: All tool functions use @mcp.tool and @mcp_auth_hook decorators for registration and auth (no more add_tool); DAOs/commands imported inside function body
DAO wrapper removed: The old dao wrapper has been removed for clarity and maintainability
Auth/RBAC/logging hooks: Stubbed in auth.py and middleware.py, admin mode by default, ready for extension
Extension points: Documented and ready for Preset/enterprise
Core actions implemented:
- list_dashboards, list_datasets (now returns columns and metrics), list_charts
- get_dashboard_info, get_dataset_info (now returns columns and metrics), get_chart_info
- get_dashboard_available_filters, get_dataset_available_filters, get_chart_available_filters
- create_chart_simple (PoC for mutation)
- create_chart (advanced ECharts chart creation, supports stack, area, smooth, show_value, color_scheme, legend_type, legend_orientation, tooltip_sorting, y_axis_format, y_axis_bounds, x_axis_time_format, rich_tooltip, extra_options)
- get_superset_instance_info
Tests: Unit and integration tests for all core tools, with improved coverage and best practices. Dataset and chart tools now have tests verifying advanced ECharts options and extra_options are included in responses.
Docs: Architecture, schemas, and dev guides up to date
Tool module reorganization: Modules have been reorganized for clarity and maintainability
Chart creation tool modeling: Progress on modeling chart creation tool input parameters for flexibility and LLM-friendliness

What’s Next / Planned

Primary focus: chart_create tool
- Design and implement a robust chart creation tool
- Explore how prompt engineering can guide chart creation (e.g., mapping user intent to chart config)
- Integrate prompt injection prevention from the start
- Support creating a chart and navigating the user to Explore with the settings (without saving the chart)
- Investigate backend rendering of charts for previews or agent feedback
- Add mutation tools for iterating on charts, dashboards, datasets, etc. (e.g., update, refine, clone)
More advanced mutations (create/update/delete)
Navigation tools (e.g., generate_explore_link, open_sql_lab_with_context)
Full RBAC, impersonation, logging (stubbed only in Phase 1)
Preset/enterprise integration (hooks ready)

Summary Table

SoW Objective/Deliverable	Status (Phase 1)
Standalone MCP service, config flag, CLI	✅ Delivered
Modular, typed schemas	✅ Delivered
Abstracted, reusable list/info/filter tools	✅ Delivered
3+ core actions (list/count)	✅ Delivered
Mutations (create/update)	🟡 PoC for chart creation
Navigation actions	🟡 Planned
Auth/RBAC hooks (stubbed)	✅ Delivered
Documentation & dev guide	✅ Delivered
Preset extension points	✅ Delivered
Demo script/notebook	🟡 Not in OSS, easy to run/tests

Reference

SIP-171: MCP Service Proposal

4.9 KiB Raw Blame History Unescape Escape