start_dataproc_clusterLABSCO SUMMARY
Profiles carry the parameters, so most calls take a fraction of what the Dataproc API wants.
Project and region are injected from config and cluster shape comes from a profile, which is why creating a cluster or listing them is a one-line request rather than a wall of arguments. Job cancellation is worth knowing about on its own — a runaway Spark job needs only its job ID to stop.
What it is
An MCP server for Google Cloud Dataproc that injects project and region defaults, ships cluster profiles, and covers cluster lifecycle, job submission and analytics.
What you get
- Cluster lifecycle: start_dataproc_cluster, create_cluster_from_yaml, create_cluster_from_profile, list_clusters, get_cluster, delete_cluster
- list_tracked_clusters for clusters this server created, and get_zeppelin_url for the notebook UI
- Jobs: submit_dataproc_job for Spark, PySpark and Presto, plus submit_hive_query, cancel_dataproc_job, get_job_status, get_job_results
- Hive queries tracked separately: get_query_status and get_query_results, with pagination
- Profiles and stored data: list_profiles, get_profile, query_cluster_data
- Analytics: check_active_jobs, get_cluster_insights, get_job_analytics, query_knowledge
- 8 built-in cluster profiles, so a cluster comes up without hand-writing the config
Requirements
Run it with npx @dipseth/dataproc-mcp-server@latest, or install globally as dataproc-mcp-server. DATAPROC_CONFIG_PATH points at your config. Authentication is service-account impersonation, a direct service-account key, application default credentials, or a hybrid with fallbacks. Version 4.7.0.
Setup effort
One command plus credentials — npx @dipseth/dataproc-mcp-server@latest, then supply credentials
USE CASES
01
Bringing up a cluster without writing the config
Pick a profile and create from it; project and region come from your defaults.
02
Running a Spark job and watching it
Submit, poll the status, then pull the output when it finishes.
03
Stopping a job that is burning money
Cancel it by job ID, without opening the console.
04
Working out why jobs keep failing
Ask for success rates and error patterns rather than reading logs one run at a time.
22 tools
Listed in project docs. Your client may show a different list.
create_cluster_from_yamlcreate_cluster_from_profilelist_clusterslist_tracked_clustersget_clusterdelete_clusterget_zeppelin_urlsubmit_hive_querysubmit_dataproc_jobcancel_dataproc_jobget_job_statusget_job_resultsget_query_statusget_query_resultslist_profilesget_profilequery_cluster_datacheck_active_jobsget_cluster_insightsget_job_analyticsquery_knowledgeBEFORE YOU INSTALL
What to know first
Tools you get
22 tools named in the project's documentation.
Requires
- npx on PATH
- GOOGLE_APPLICATION_CREDENTIALS in your environment
Where it runs
Both documented clients run the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
npx @dipseth/dataproc-mcp-server@latestMCP Clients
Roo Code
ROO CODE · JSON
{
"mcpServers": {
"dataproc": {
"command": "npx",
"args": [
"@dipseth/dataproc-mcp-server@latest"
],
"env": {
"LOG_LEVEL": "info"
},
"alwaysAllow": []
}
}
}The configuration on this page does not set GOOGLE_APPLICATION_CREDENTIALS; the project's instructions set it with an export line. An app opened from its icon or dock does not see variables exported in a terminal, so to supply GOOGLE_APPLICATION_CREDENTIALS, use the block's "env" section.
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"dataproc": {
"command": "npx",
"args": [
"@dipseth/dataproc-mcp-server@latest"
],
"env": {
"LOG_LEVEL": "info"
}
}
}
}The configuration on this page does not set GOOGLE_APPLICATION_CREDENTIALS; the project's instructions set it with an export line. An app opened from its icon or dock does not see variables exported in a terminal, so to supply GOOGLE_APPLICATION_CREDENTIALS, use the block's "env" section.
