Tabularis: Open-Source Desktop SQL Workspace
Open-source desktop SQL workspace for PostgreSQL, MySQL, SQLite and 15+ other databases, with a built-in MCP server for AI agents.
- Source repo
- TabularisDB/tabularis
- Stars
- ★ 5.2k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 47/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformChatGPT · Codex · Claude Code · Claude.ai
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A data analyst uses the visual query builder to drag-and-drop tables and create aggregate queries without writing SQL.
- Not a fit if
- Teams needing dozens of database drivers; only 3 ship built in
- Users unwilling to configure external AI provider API keys
- Users needing drivers still in development, like SQL Server
- Source review
- 47/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Tabularis is an open-source desktop SQL workspace supporting PostgreSQL, MySQL/MariaDB, SQLite, and 15+ more databases like DuckDB, ClickHouse, Redis, and Firestore via plugins. Built with Rust, Tauri v2, and React, it offers a SQL editor, SQL notebooks, a visual query builder, and visual EXPLAIN. Its built-in MCP server lets AI agents like Claude, Cursor, and Devin read your schema and run queries directly from their interfaces. It also supports plugins written in any language over JSON-RPC and local AI models via Ollama. The app UI is available in multiple languages and is distributed via WinGet, Homebrew, Snap, and more.
Tabularis runs as a desktop application, providing SQL editing via the Monaco editor with multi-statement execution, query splitting, and AI assist. SQL notebooks combine Markdown and SQL cells with cross-cell variables and charts. The visual query builder uses ReactFlow to drag-and-drop tables and generate SQL. Visual EXPLAIN renders execution plans as interactive graphs. The data grid supports inline editing and JSON/JSONB cells. The MCP server runs with 'tabularis --mcp' and exposes tools to AI agents: list_connections, list_databases, list_tables, describe_table, and run_query. Plugins communicate over JSON-RPC 2.0 via stdin/stdout, extending support to databases like MongoDB and DuckDB.
- A data analyst uses the visual query builder to drag-and-drop tables and create aggregate queries without writing SQL.
- A developer uses SQL notebooks to document queries and share results with charts and parameters.
- An AI agent like Claude reads database schema and executes queries through the MCP server to retrieve data.
- A DBA uses visual EXPLAIN to analyze and optimize PostgreSQL query performance.
- A team writes custom database drivers as plugins and integrates them into Tabularis.
How do you install or deploy this agent?
On Windows: winget install Debba.Tabularis. On macOS: brew tap TabularisDB/tabularis && brew install --cask tabularis. On Linux: sudo snap install tabularis or download the AppImage from releases.
How do you use this agent?
Launch Tabularis, create a database connection (PostgreSQL, MySQL, SQLite, etc.). In the SQL editor, type and run queries, or use SQL notebooks for multi-cell documents. For AI integration, run 'tabularis --mcp' and install the config for Claude Desktop, Cursor, or Devin in Settings. AI agents will be able to list connections, list databases, list tables, describe tables, and run queries.
What are this agent's strengths and limitations?
- Built-in MCP server for direct integration with AI agents like Claude, Cursor, and Devin
- Supports 15+ databases with a plugin system for any database
- Advanced features: SQL notebooks, visual query builder, and visual EXPLAIN
- Local AI models via Ollama, no API key required for privacy
- Fewer built-in databases compared to DBeaver (3 built-in + 16 official plugins)
- Some plugins are 'claimed' or 'scoped', possibly not fully complete or available
- AI features require configuration of external API providers (unless using Ollama)
- Plugin system requires technical knowledge to write custom plugins
How does this agent compare with similar options?
Compared to DBeaver CE, Tabularis offers a built-in MCP server and SQL notebooks, while DBeaver supports more databases. Compared to TablePlus, Tabularis is open-source, whereas TablePlus is commercial.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Tabularis: Open-Source Desktop SQL Workspace This agent | 47 · Major gaps | Desktop appFree + model costs | ★ 5.2k | today | TypeScript | ChatGPT · Codex · Claude Code · Claude.ai |
| WhoDB | 51 · Major gaps | Self-hosted serviceFreemium | ★ 5k | today | Go | — |
| Mnemosyne | 59 · Major gaps | CLIFree | ★ 3.5k | today | Python | Codex · Claude Code |
| Nocturne Memory: Long-Term Memory Server for MCP Agents | 56 · Major gaps | MCP serverFree | ★ 1.4k | 26d ago | Python | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: MCP server provides read-only mode and write-approval gate (SECURITY.md), but implementation details are not provided; user confirmation mechanism exists but not detailed; data flow transparency limited, no explanation of data transmission; sensitive data handling: passwords stored in system keychain, but encryption details not specified; dependency security: lockfile and pinned versions used, but no vulnerability scanning evidence; external effects: MCP tools can execute queries, but permission control not specified; rollback mechanism not mentioned; source attribution: clear copyright and author info. Deductions: lack of implementation details and verification evidence.
Evidence shows: README and SECURITY.md descriptions consistent, but no test results provided; dependency availability: pnpm and lockfile used, but no guarantee of dependency availability; failure messages: no error handling examples. Deductions: lack of test evidence and error handling documentation.
Evidence shows: target users clear (database developers), scenarios diverse (SQL editing, MCP integration); capability boundaries clear (supports multiple databases, plugin system); trigger precision: MCP tool names clear, but trigger conditions not specified; environment fit: supports multiple platforms and installation methods. Deductions: insufficient trigger precision and boundary details.
Evidence shows: information architecture clear (README, SECURITY, plugin guide); install notes detailed (multi-platform commands); naming stability: version numbers clear, but API stability not specified; examples and FAQ: usage examples present, but no FAQ; known limitations: not explicitly listed; license: Apache-2.0, complete; versioning changelog exists (CHANGELOG.md); maintenance responsibility clear (maintainer contact info). Deductions: missing FAQ and explicit limitations.
Evidence shows: output usability: multiple output formats (CSV, JSON); marginal value: comparison with competitors highlights features; cost-benefit: open-source free, but no performance data. Deductions: lack of performance evidence.
Evidence shows: claims traceable (README mentions features, but no implementation links); cross-source corroboration: multiple language READMEs, but no independent verification; fact-inference separation: some claims clear, but not distinguished. Deductions: lack of independent verification and clear separation.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- MCP safety mechanisms (read-only mode, write approval) are only declared in SECURITY.md, no implementation details provided; actual code needs verification.
- Dependency security lacks vulnerability scanning evidence; recommend checking dependency versions and known vulnerabilities.
- Rollback mechanism not mentioned; for a database tool, recommend clarifying undo or recovery options for data modifications.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.