Labsco
MCP SERVER · OFFICIAL PROJECT

Deephaven MCP

by deephaven

Query live Deephaven tables, run scripts on a session, and manage enterprise persistent queries.

Data Platform: Pipelines, Warehousing, BI & Governance
Summary
Batch PQ operations report per item instead of failing the lot.

pq_start, pq_stop, pq_restart and pq_delete each take a list and run in parallel under an operator-set concurrency cap, so one bad query does not sink a restart of twenty. dhcli describes itself at runtime through its own agent manifest, which is how an agent keeps up with a command surface the project says is still moving.

What it is

Deephaven's own server. One multiplexed process hosts every configured Community session and Enterprise system; a separate hosted documentation server answers questions about Deephaven itself.

What you get
  • Configured sessions and systems listed, and an enterprise system's health read — list_systems, enterprise_systems_status
  • Sessions created and deleted on either side, with credentials retrievable subject to policy — session_community_create, session_community_delete, session_community_credentials, session_enterprise_create, session_enterprise_delete
  • Tables discovered and read on any session: names, schema and data with formatting options — sessions_list, session_details, session_tables_list, session_table_schema, session_table_data
  • Python or Groovy executed on a session, and its installed packages listed — session_script_run, session_pip_list
  • Persistent queries across their whole lifecycle, with start, stop, restart and delete taking a list of ids, running in parallel and reporting per item — pq_list, pq_details, pq_create, pq_modify, pq_start, pq_stop, pq_restart, pq_delete, pq_name_to_id
  • The enterprise catalog browsed at table and namespace level, with schema and sample reads — catalog_tables_list, catalog_namespaces_list, catalog_table_schema, catalog_table_sample
  • A hosted documentation server at https://deephaven-mcp-docs-prod.dhc-demo.deephaven.io/mcp answering natural-language Deephaven questions
Requirements

Deephaven-mcp from PyPI provides three commands: dhcli, dh-mcp-systems-server and dh-mcp-docs-server. Configure with dhcli config init rather than editing JSON by hand. The documentation server is hosted and speaks streamable HTTP only, so stdio-only clients reach it through mcp-proxy. Configuration changes require a server restart.

Setup effort

One command — uv tool install --python-preference managed "deephaven-mcp"