Superset MCP Server
Allows AI agents to create and manage dashboards, charts, and filters in Apache Superset, using datasets from databases like Snowflake.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Superset MCP Server create a dashboard for monthly revenue from the sales dataset"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Superset MCP Server
MCP server that exposes Apache Superset as tools so the AI can build dashboards from Cursor: create dashboards, add charts from datasets (e.g. Snowflake views), and set filters from your instructions.
Sharing with the team
Best approach: Each colleague does a one-time setup on their machine with their own Superset credentials. No credentials are stored in the repo.
Each person: Follows TEAM_SETUP.md once: clone →
pip install -e .→ get their own auth (see GET_TOKEN.md) → add Superset MCP in Cursor with their path and credentials → reload MCP.
Related MCP server: metabase-mcp
Prerequisites
Python 3.10+
A running Superset instance with API enabled
In Superset: at least one database (e.g. Snowflake) and datasets (tables/views) that you want to use in dashboards
Setup
1. Install dependencies
From the directory that contains mcp_superset (the folder can live anywhere):
cd <path-to-mcp_superset>
pip install -e .
# or with uv:
uv pip install -e .2. Environment variables
Use one of: (A) session cookie, (B) access token, or (C) username/password. Set in Cursor MCP config or your shell:
Variable | Required | Description |
| Yes | Base URL of Superset (e.g. |
Option A – Session cookie (browser / Google login, no JWT) | ||
| Yes* | Cookie string, e.g. |
Option B – Access token | ||
| Yes* | JWT from browser. Optional: |
Option C – Username/password | ||
| Yes | API user (e.g. |
| Yes | Password for that user |
| No | Auth provider; default |
Session cookie (when you only have cookie, no Authorization header):
See GET_TOKEN.md: log in to Superset, F12 -> Application -> Cookies -> your Superset URL -> copy the session cookie value, then set SUPERSET_SESSION_COOKIE=session=<paste value>. Session expires when you close the browser or after some time; get a fresh cookie when you get 401s.
Do not commit credentials. Use Cursor’s MCP env or a local .env that is gitignored.
3. Add the server in Cursor
Open Cursor Settings → Features → MCP.
Click Add new MCP server.
Choose Run a script / command (stdio).
Configure:
Option A – Use your Python (recommended)
Command:
python(or the full path to your Python / venv, e.g.<path-to-mcp_superset>\.venv\Scripts\python.exe)Arguments:
-m mcp_superset.serverWorking directory:
Full path to the mcp_superset folder (e.g.c:\Bio\cursor_projects\mcp_superset)Env (add here or in Cursor MCP env):
SUPERSET_URL,SUPERSET_USERNAME,SUPERSET_PASSWORD
Option B – Global install
If you installed the package globally:
Command:
mcp-server-supersetEnv: same as above.
Option C – JSON config (Cursor MCP)
If your Cursor MCP is configured via JSON, add something like:
{
"mcpServers": {
"superset": {
"command": "python",
"args": ["-m", "mcp_superset.server"],
"cwd": "C:\\path\\to\\mcp_superset",
"env": {
"SUPERSET_URL": "https://superset.yourcompany.com",
"SUPERSET_USERNAME": "your_user",
"SUPERSET_PASSWORD": "your_password"
}
}
}
}Replace C:\\path\\to\\mcp_superset with the actual path where you placed the folder.
Restart Cursor or reload MCP after adding the server.
Tools exposed to the AI
Tool | Purpose |
| List Superset databases (e.g. Snowflake connection) |
| List datasets; optional |
| Get dataset by id (columns, metrics) for building charts |
| List dashboards; optional |
| Get dashboard by id or slug (layout, metadata, filters) |
| Create empty dashboard; then add charts and filters |
| Update dashboard (title, slug, published) |
| Delete a dashboard by id |
| Set native filters (JSON array of filter config) |
| Add chart to dashboard with position (x, y, width, height) |
| List charts; optional |
| Get chart by id |
| Create chart (dataset_id, viz_type, slice_name, params JSON) |
| Update chart (slice_name, params, description) |
| Delete a chart by id |
| List charts on a dashboard |
Workflow: Snowflake views → Superset dashboard
You tell the AI what you want: e.g. “Dashboard for view X, filter by date and region, bar chart and table.”
AI uses Snowflake MCP to inspect views/tables (e.g.
list_objects,run_snowflake_query).AI uses Superset MCP to:
superset_list_datasetsto find the dataset that points at that viewsuperset_get_datasetto see columnssuperset_create_dashboardand thensuperset_create_chartfor each chartsuperset_add_chart_to_dashboardto place themsuperset_update_dashboard_filtersto add the filters you asked for
You open the dashboard in Superset and refine if needed.
Native filters (dashboard filters)
superset_update_dashboard_filters takes a JSON string that is an array of filter objects. Each object typically has:
id: unique string id for the filtername: label shown in the UIfilterType: e.g.filter_select,filter_time,filter_timegraintargets: which charts/columns the filter applies todefaultDataMask: default valuescope: scope of the filter
The AI can build this from your instructions (e.g. “add a date range and a region dropdown”) by following Superset’s native filter schema.
Chart params
For superset_create_chart, params is a JSON string object. Contents depend on viz_type, for example:
table:
metrics,groupby,order_desc,row_limit, etc.big_number:
metric,compare_lag, etc.line / bar:
metrics,groupby,time_range,order_desc, etc.
The AI should use superset_get_dataset to see available columns/metrics and build valid params.
Run the server locally (optional)
cd mcp_superset
set SUPERSET_URL=https://...
set SUPERSET_USERNAME=admin
set SUPERSET_PASSWORD=...
python -m mcp_superset.serverThe server uses stdio; Cursor will start it automatically when the tools are used.
Available Tools
17 toolssuperset_add_chart_to_dashboardA
Add an existing chart to a dashboard at the given grid position (x, y, width, height).
Inputs: chart_id (required), dashboard_id (required), x, y, +2 more. Returns: result
superset_create_chartB
Create a new chart. params must be a JSON string: object with keys like metrics, groupby, order_desc, row_limit, time_range, etc. depending on viz_type (e.g. table, big_number, line, bar, pie).
Inputs: params (required), viz_type (required), dataset_id (required), slice_name (required), +1 more. Returns: result
superset_create_dashboardA
Create a new empty dashboard. Returns the new dashboard with id. Then add charts and filters.
Inputs: dashboard_title (required), slug, published. Returns: result
superset_delete_chartA
Delete a chart by id. Removing it from a dashboard does not delete the chart; use this to delete it from Superset.
Inputs: chart_id (required). Returns: result
superset_delete_dashboardA
Delete a dashboard by id. This does not delete the charts on it; they remain in Superset.
Inputs: dashboard_id (required). Returns: result
superset_get_chartB
Get a chart by id. Returns viz_type, params, datasource, etc.
Inputs: chart_id (required). Returns: result
superset_get_dashboardA
Get a dashboard by id or slug. Returns full detail including position_json and json_metadata (filters).
Inputs: id_or_slug (required). Returns: result
superset_get_dashboard_chartsB
Get all charts that belong to a dashboard.
Inputs: dashboard_id_or_slug (required). Returns: result
superset_get_datasetA
Get a dataset by id, including columns and metrics. Use to see available columns before creating charts.
Inputs: dataset_id (required). Returns: result
superset_list_all_datasetsA
List all datasets in Superset across all pages. Use when you need the full list (e.g. to find a dataset by name like vw_bioverse_performance_report).
Inputs: page_size. Returns: result
superset_list_chartsA
List charts. Optionally search by name.
Inputs: search. Returns: result
superset_list_dashboardsA
List dashboards. Optionally search by title.
Inputs: search. Returns: result
superset_list_databasesA
List all databases configured in Superset (e.g. Snowflake). Use to find database_id for listing datasets.
Inputs: no named parameters. Returns: result
superset_list_datasetsA
List datasets (tables/views) in Superset. Optionally filter by database_id or search by name. Use to find dataset_id for creating charts.
Inputs: search, database_id. Returns: result
superset_update_chartA
Update an existing chart. Pass only the fields you want to change. params must be a JSON string if provided (same shape as for create_chart).
Inputs: chart_id (required), params, slice_name, description. Returns: result
superset_update_dashboardA
Update an existing dashboard. Pass only the fields you want to change (e.g. title, slug, published).
Inputs: dashboard_id (required), slug, published, dashboard_title. Returns: result
superset_update_dashboard_filtersA
Update native (dashboard-level) filters. Pass native_filter_configuration as a JSON string: list of filter objects with id, name, filterType, targets, defaultDataMask, scope, etc. Replaces all existing native filters.
Inputs: dashboard_id (required), native_filter_configuration (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 17 tools
Most tools target distinct resources and actions. The only potential confusion is between superset_list_datasets and superset_list_all_datasets, but their descriptions clearly differentiate (filtered/search vs all pages). superset_get_dashboard_charts also overlaps slightly with superset_list_charts, but the dashboard scoping is clear.
All tools follow a consistent `superset_verb_noun` pattern with snake_case. The verb is always placed immediately after the prefix, and nouns are lowercase (e.g., superset_list_databases, superset_create_dashboard). The only minor deviation is superset_list_all_datasets, but it's still readable and predictable.
17 tools is slightly above the typical sweet spot, but each tool addresses a distinct operation on the core entities (databases, datasets, dashboards, charts). There are no trivial or redundant tools, so the count feels justified for full chart/dashboard lifecycle management.
Charts and dashboards have complete CRUD coverage plus extra operations (filters, adding charts). However, databases and datasets only support listing/getting, with no create, update, or delete. Additionally, there is no way to remove a chart from a dashboard without deleting it, which is a notable gap.
Related MCP Connectors
Connect AI assistants to Preset to explore data, charts, and dashboards in authorized workspaces.
Company and industrial data for AI assistants: datasets, SQL, dashboards, alarms, with approvals.
Agent-Native Amplitude/Mixpanel - connect data sources, prompt for charts
Lets AI agents analyze API metrics, request logs, and traces, and manage apps and alerts.
Related MCP Servers
- AlicenseBqualityDmaintenanceMCP server enabling AI agents to programmatically connect to and control Apache Superset instances, allowing users to manage dashboards, charts, databases, datasets, and run SQL queries through natural language interactions.60195MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to interact with Metabase for database operations, SQL queries, dashboard management, and analytics automation.28MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to manage Redash queries, dashboards, visualizations, and data sources via natural language.1512 npmMIT
- AlicenseBqualityDmaintenanceConnects LLM clients to Apache Superset for querying charts, dashboards, datasets, databases, and executing SQL.121GPL 3.0