Tabularis: a desktop SQL workspace with an MCP server for your coding agent

A workspace, with a comparison table attached

Tabularis is an open-source desktop SQL workspace, and its why section is a comparison table against DBeaver CE, TablePlus and Beekeeper Studio that doubles as a feature list. The unique-marked rows are SQL notebooks, SQL plus Markdown cells with cross-cell variables and charts, a built-in MCP server for AI agents, plugins in any language over JSON-RPC on stdio, and AI text-to-SQL with local models through Ollama, each checked only for Tabularis. Visual EXPLAIN with interactive plan graphs it shares with DBeaver, and database coverage it concedes, three built-in drivers plus 21 official plugins against DBeaver's hundred plus. The footnote carries the honesty, comparison as of June 2026, features in other tools may have changed, and if you need dozens of drivers, use DBeaver, Tabularis focuses on doing a few databases well. The origin story section promised by the table of contents and the plugin registry link on the website complete the picture of a project that documents its own reasoning rather than only its features.

Three built-in, twenty-one plugins, one registry

PostgreSQL, MySQL/MariaDB and SQLite ship built in, and everything else is a plugin, with the built-in PostgreSQL driver deprecated in favor of a PostgreSQL plugin the app installs automatically. The shipped plugin list spans the analytics and NoSQL worlds, ClickHouse, DuckDB, Cloudflare D1 in two flavors, DynamoDB, Elasticsearch, Firestore, IBM Db2 and Informix, LibSQL and Turso, MongoDB with Atlas, Oracle, Redis in Go and Rust implementations, SQL Server, plus the unusual CSV Folder, Google Sheets and HackerNews drivers. Beyond the shipped set, a status ladder extends, Google BigQuery, Meilisearch and Amazon Redshift claimed, CockroachDB, TiDB and Snowflake scoped or coming soon, and Cassandra, Etcd, Firebird, ScyllaDB, SurrealDB and Trino open, statuses installable from the plugin registry on the website.

Eleven readmes and one install matrix

The README ships in eleven languages, English through Chinese, Japanese, Korean, Russian, Tagalog and Portuguese, and the app UI itself covers ten of them, an investment matching a desktop tool's global audience. Installation covers every desktop channel, winget on Windows, brew cask on macOS, snap, flatpak, AUR bin and winstall entries, plus direct installers for Windows exe, both Intel and Apple Silicon dmg, AppImage, deb and rpm. The download section shows the three main package manager commands:

winget install Debba.Tabularis                                   # Windows
brew install --cask tabularis  # macOS
sudo snap install tabularis                                      # Linux

The version pinned in the links, 0.25.0, is the release the installers carry, while nightly builds nearly every day, three between September 28 and 29 of 2026, show the development pace, with the repository pushed the day before this writing.

The MCP server: agents inside your SQL app

The built-in MCP server is the feature the description leads with, letting Claude, Cursor and Devin, formerly Windsurf, read your schema and run queries in the same app you already use. The architecture inverts the usual AI assistant shape, instead of an agent with its own database connections, the agent connects to the workspace where the connections, credentials and query history already live, so permissions and context stay in one place. The table of contents pairs the MCP section with optional AI features, the text-to-SQL capability that works with local models through Ollama, keeping schema data on the machine when cloud assistants are unwanted. For teams already working with coding agents, the workspace becomes the database surface those agents act through rather than a parallel tool. The MCP naming matters for evaluation, Model Context Protocol is the standard Claude, Cursor and Devin speak for tool access, so a server inside the desktop app means the agent operates through the same permissions and connection pool the human uses, rather than receiving its own credentials.

The feature list, from grid to graph

The feature sections cover the workspace's daily surface. Connection management and a database explorer organize the left side, the SQL editor with a visual query builder handles writing, and SQL notebooks mix SQL and Markdown cells with cross-cell variables and charts, the format the demo directory showcases with a notebook file and a video. Visual EXPLAIN renders query plans as interactive graphs, the diagnosis view the comparison table highlights, and the data grid handles editing results. Keyboard shortcuts, logging and a plugin system complete the list, with the plugin API published as its own package and a create-plugin scaffold, so extending the app is a documented path rather than a fork.

Tauri, Monaco, and a dual-language test suite

The tech stack splits between TypeScript and Rust, the src directory building with Vite and the src-tauri directory holding the Tauri shell, tested with cargo test. Monaco, the editor engine behind VS Code, appears in the dependencies, driving the SQL editor, and the package structure isolates the explain visualization and the plugin API as separately buildable packages with their own typecheck and smoke scripts. The test surface spans both languages, vitest with coverage on the TypeScript side and a dedicated theme contrast test, and the lint script checks theme tokens with a custom script, the discipline of an app whose theming is a feature. Release automation syncs versions across the README files, changelog and Tauri config in one version script. The theme token checker and contrast test imply a design system maintained in code, colors defined once and verified for accessibility across themes, which matters in an app users stare at for hours.

A demo, a roadmap in JSON, and an origin story

The repository carries unusual documentation furniture, a demo directory with a docker-compose file, a connections file, a perf SQL generator and a showcase notebook, so a reviewer can reproduce the showcase environment rather than watching a video alone. The roadmap lives as roadmap.json, generated by a script, making the plan machine-readable, and the table of contents promises an origin story section, the narrative most tools leave to a blog post. Community surfaces include Discord, Bluesky, Mastodon and X links in the package metadata, and the sponsors file sits beside the license, Apache-2.0. The BUGS file at the repository root, a plain text of known issues, is the kind of honesty small projects adopt before issue trackers feel necessary.

Editorial conclusion

Use Tabularis when day to day database work spans a handful of engines, its three built-in drivers plus plugin registry, and when the SQL notebook format or agent integration through MCP would change your workflow, since those are the features the comparison table marks as unique. If you need dozens of drivers, the project's own advice is to use DBeaver. Before adopting, note the maturity signal, version 0.25 with nightly builds landing daily, verify your platform's install channel from winget through brew to snap, flatpak, AUR and AppImage, and read the AI features section before enabling text-to-SQL, since local model support through Ollama is the privacy-preserving option.

Frequently asked questions

Official sources

1. License: Apache-2.0
2. Project website
3. README
4. Releases
5. TabularisDB/tabularis on GitHub