Query ClickHouse over its HTTP interface with per-tab sessions, browse databases down to engine families, key columns and codecs, watch merges, mutations and replication in a performance pane, and edit rows only after the server has proved which row a mutation will touch. It's a native ClickHouse GUI for Mac and Windows that also speaks MySQL, PostgreSQL, Redshift, DynamoDB, Redis, MongoDB and Elasticsearch, with no account and no telemetry.
brew install --cask shubhesh07/db-connect/db-connect
curl -fsSL https://github.com/shubhesh07/db-connect/releases/latest/download/install.sh | bash
Write ClickHouse SQL as you would in clickhouse-client; autocomplete knows ClickHouse's keywords and functions, and Explain shows the plan with the parts and granules each index pruned (or Estimate, Pipeline and Syntax):
SELECT toStartOfHour(ts) AS hour, count() AS events, uniqExact(user_id) AS users
FROM events
WHERE ts >= now() - INTERVAL 1 DAY
GROUP BY hour
ORDER BY hour
Thirty built-in snippets go straight to the system tables: running queries, slow shapes from query_log, parts per partition, merges, mutations, replication lag, memory, dictionaries and disks. For example, the one that finds tables getting too many small inserts:
SELECT database, table, partition, count() AS parts, sum(rows) AS rows
FROM system.parts
WHERE active
GROUP BY database, table, partition
ORDER BY parts DESC
LIMIT 20
Self-managed ClickHouse or ClickHouse Cloud over the HTTP interface: port 8123, or 8443 with TLS and an optional CA file, username and password, and an SSH tunnel through a bastion. Version, timezone and user show on connect. Cloud hosts are recognised and the form flips the default port with TLS.
Databases, then tables grouped by engine family: MergeTree, Replicated, Distributed, Log, Memory, dictionaries, views and materialized views with the table they feed. Columns carry primary, sorting and partition key badges, default kinds and codecs; skip indexes have their own folder; Users & roles opens a grants viewer. Right-click a table for Query, DDL, Describe, Parts, Column sizes, Running queries, Export to file or Import file.
Each tab has its own server session, so USE, SET and temporary tables persist between runs. 64-bit integers and decimals arrive exact, the toolbar LIMIT is placed before SETTINGS and FORMAT, and Cancel issues KILL QUERY.
Opt in per connection, then edit cells, remove rows and add rows in the grid. Before anything runs, every edited row is matched on the server: it must match exactly one row, the dialog shows the exact mutation and how many parts it rewrites, and Apply stays disabled until every change is safe. Values are bound as typed query parameters, never spliced into SQL.
Memory, load, CPU, cache hit rates, merge pool, delayed inserts, parts per partition and replica delay at a glance. Running queries with Kill, slow query shapes from system.query_log with where they were fired from, merges and mutations with Kill, replication, parts and disks, errors and dictionaries. A cluster picker switches every list to clusterAllReplicas.
Export CSV, JSON Lines or Parquet streamed straight from the server with FORMAT, with progress. Import CSV, TSV, JSON Lines or Parquet into a table from the native file picker. Watch re-runs a read on an interval and alerts on a condition.
Read-only connections send readonly=2 and allow_ddl=0 on every request, so the server enforces them. Production connections confirm every write: inserts, mutations, OPTIMIZE, DDL, partition operations, SYSTEM, KILL, GRANT and REVOKE.
The local MCP server exposes list_databases, list_tables, describe_table, run_query, explain_query, active_sessions, slow_queries and cluster_health for allow-listed ClickHouse connections; without a write grant queries run with readonly=2.
The Play UI that ships with every ClickHouse server is free and needs nothing installed, but it is one query box per browser tab, with no schema explorer, history or safety rails. Tabix is an open-source web client with a schema tree and charts. DBeaver and DataGrip reach ClickHouse through the JDBC driver and are capable general-purpose IDEs, but they do not read the system tables into a performance view or check what a mutation will touch before running it. DB Connect is for engineers who query ClickHouse next to the transactional databases, caches and search clusters behind it: one desktop app with every connection, per-tab sessions, a performance pane, grid edits behind a server-side dry run, and a local MCP server for AI tools. Pick by workflow, not by feature count, and check each tool's current docs, since all of them change often.
Corrections are welcome: open an issue.
Yes, since v3.8.0. It connects over the ClickHouse HTTP interface (port 8123, or 8443 with TLS) to self-managed servers and ClickHouse Cloud. The same app also covers MySQL, PostgreSQL, Redshift, DynamoDB, Redis, MongoDB and Elasticsearch/OpenSearch.
Yes. DB Connect is free for personal and commercial use, runs natively on macOS (Intel and Apple Silicon) and Windows x64, and needs no account. Install it with brew install --cask shubhesh07/db-connect/db-connect or download the DMG.
Yes, once you turn on "Allow grid edits" for the connection. ClickHouse has no unique keys and a mutation reports no row count and cannot be rolled back, so DB Connect runs a dry run first: each edited row is matched on the server and must match exactly one row, the dialog shows the exact statement and how many parts it rewrites, and Apply stays disabled until every change is safe. Identical twin rows and rows that vanished since the query are refused.
Yes. Use the service host (it ends in .clickhouse.cloud) on port 8443 with TLS, and the user and password from the Cloud console. The form flips the default port when you toggle TLS and recognises Cloud hosts.
Read-only connections send readonly=2 and allow_ddl=0 on every request, so the server itself refuses writes and DDL. Production-tagged connections ask before every write, including INSERT, mutations, OPTIMIZE, DDL, partition operations, SYSTEM and KILL, and every grid edit is shown as a dry run first. Credentials are encrypted with the key in your OS keychain.
Yes, through DB Connect's local MCP server. It's off by default, listens on 127.0.0.1, needs a bearer token, and only reaches connections you allow-list. list_databases, list_tables, describe_table, run_query, explain_query, active_sessions, slow_queries and cluster_health work on ClickHouse connections; without a write grant run_query runs with readonly=2.
macOS (Intel + Apple Silicon) and Windows x64, about 35 MB. Also available via brew install --cask shubhesh07/db-connect/db-connect.