| FMRS | 57 / 100 · C | 74 / 100 · B | 74 / 100 · B |
| Reliability | 8 / 20 | 13 / 20 | 13 / 20 |
|---|
| Security and permissions | 10 / 20 | 13 / 20 | 13 / 20 |
|---|
| Maintenance | 14 / 20 | 17 / 20 | 16 / 20 |
|---|
| Documentation | 13 / 20 | 16 / 20 | 15 / 20 |
|---|
| Setup experience | 12 / 20 | 15 / 20 | 17 / 20 |
| Best for | - Developers already using DBX desktop, web, or Docker with connections configured
- Teams working across many different database engines who want one consistent MCP access point
- Users who want a lightweight (20 MB) client without Java or Python runtime dependencies
| - Teams that want precise control over which database operations an AI can perform, rather than open arbitrary SQL execution
- Scenarios needing a unified MCP setup across multiple database engines
| - Scenarios wanting the smallest possible tool count and lowest context-window overhead for database access
- Cases needing to grant non-technical teammates safe, read-only database access
|
| Not for | - Users without an existing DBX installation or configured connections, since the MCP server itself does not manage new database credentials — it relies on connections already set up in DBX
- Users who specifically need an MCP server maintained by a database vendor (e.g. the MySQL, PostgreSQL, or MongoDB projects themselves) rather than DBX's own client wrapper
- Environments with strict no-write automation requirements where operators cannot verify the current DBX centralized MCP access-mode policy
| - Lightweight cases that just want to run a few ad-hoc SQL queries without maintaining a tools.yaml config (a simpler single-database MCP may be a better fit)
| - Scenarios needing database administration operations (creating indexes, altering schema) rather than just querying (DBHub is positioned as a query tool, not a DBA management tool)
|
| Required permissions | - Access to database connection credentials already configured in DBX
- Ability to execute SQL according to the mode selected in DBX Settings → MCP (read_only / safe_write / high_risk_write)
- DBX_WEB_URL (and DBX_WEB_PASSWORD if password login is enabled) for Web/Docker deployments
- DBX_DATA_DIR pointing to the data directory containing dbx.db for Windows portable builds
- Node.js >= 18 when run via npx (or use the Node-free precompiled native binary instead)
| - Database credentials (username/password/connection string) are supplied via env vars or config
- A tool's actual permission is whatever SQL statement is defined in tools.yaml — designed for least privilege, but misconfiguration can still over-expose access
| - Requires a database connection string (DSN), which may contain sensitive credentials
- execute_sql can run arbitrary SQL (including writes) by default — must be constrained via read-only mode or the underlying database account's own permissions
|
| Risks and side effects | - If the access mode is set to Data read/write or Full access, the AI agent can modify or delete data in connected databases
- Insufficiently reviewed AI-generated SQL could execute unintended write operations against production databases
- The legacy DBX_MCP_ALLOW_WRITES=0 environment variable only enforces read-only behavior until a central MCP policy has been saved for the first time — after that it no longer applies, so operators must track policy state
- DBX_WEB_PASSWORD is stored in plaintext in client configuration, so config files should be protected
| - If tools.yaml defines SQL statements that allow unconstrained writes or deletes, the AI could accidentally modify data
- The prebuilt toolsets (--prebuilt) favor convenience and may expose broader query capability than a specific business actually needs — use a custom tools.yaml in production
| - Under default settings, if the connected database account has write access, execute_sql could let the assistant perform an unintended write or delete — pair with a read-only database account in production
- The DSN contains a plaintext password — pay attention to the access permissions on wherever the config is stored
|
| Supported clients | Claude Code, Cursor, Windsurf | Claude Code, Gemini CLI, Zed, Antigravity | Claude Desktop, Claude Code, VS Code, Cursor, Zed |
| Tools | 0 | 0 | 2 |