Tarka

Tarka

v1.1.0 for macOS

The PostgreSQL and Snowflake query tool built for working alongside coding agents.

Claude Code, Codex and any other MCP client can connect to your databases through Tarka. Every query an agent runs appears in the app as it happens, with the agent's reasoning, the SQL, the plan and the result. Writes wait for your approval. Production data stays on your screen, not in the agent's context. Tarka has no AI of its own and needs no API keys.

Download

macOS

Universal binary (Apple Silicon + Intel), about 115 MB

Signed and notarized by Apple. Requires macOS 12 or later.

Built for agents

One MCP server, every connection

  • -Agents list your connections, run queries, explain plans, profile tables, read DDL and diff schemas or query results, all through Tarka's MCP server
  • -Queries default to EXPLAIN. Running a statement for real is an explicit choice, and DML and DDL need the agent to confirm

Guardrails you can see

  • -Writes on connections tagged production or staging wait for your Approve or Deny in the app
  • -A DELETE or UPDATE without a WHERE clause is always blocked; dangerous functions need confirmation by name
  • -Anything an agent runs without confirming executes inside a read-only transaction that is rolled back, so a write hidden in a function fails in the database instead of happening
  • -A cap on how long one agent statement may run, set by you

Your data stays yours

  • -On production and staging connections agents get column names, types and counts by default, never cell values. You see the full result in Tarka
  • -Connections, credentials, history and results live on your machine. Tarka talks to your databases directly, with no server in between

Agent sessions as tabs

  • -Each agent investigation is a tab with a live timeline: the agent's reasoning, each query, its plan summary or error, and the rows
  • -Agents work in the background without stealing your tab, editor or cursor

And a complete query tool

Query editor

  • -Syntax highlighting, autocomplete and real-time error detection
  • -Multi-tab interface with drag-and-drop reordering; run a whole script, one statement or a selection, and cancel what runs long
  • -Saved query templates, searchable history, SQL formatting, and a focus mode that gives the editor and results the full window

Results and editing

  • -Sortable, filterable results that stay fast at a million rows
  • -Edit cells, add and delete rows, then review the generated SQL before it is committed
  • -Export to CSV, JSON, SQL INSERT statements or Markdown; import a CSV into a table
  • -Pretty-printed JSON, formatted numbers, clear NULLs, timezone conversion for timestamps, and charts for numeric columns

Schema browser and table inspector

  • -Tables, views, materialized views, columns, indexes, constraints and partitions, with fuzzy search across all of them
  • -A table inspector with columns, indexes, constraints, DDL, dependencies, storage breakdown and vacuum stats
  • -Smart Lookup reads a table's indexes and builds an efficient point-lookup query for you

Query plans

  • -EXPLAIN ANALYZE as a plain-language summary, an interactive flow diagram, a flame graph and a collapsible tree
  • -Rows scanned versus planned, loop counts, timing and buffers per node, with bottlenecks highlighted

Connections

  • -PostgreSQL and Snowflake, with color tags for local, dev, staging and production that also drive the agent guardrails
  • -SSL/TLS with certificate files in every mode up to verify-full, and SSH tunnels with password or key authentication
  • -Saved passwords are encrypted with your operating system's secure storage