A checkpoint before your agent runs SQL.
Run the local MCP server in your linked project. SQL through run_sql gets a checkpoint before it runs, and a record in Cloud. The database URL, SQL and row data stay local. An agent using the raw URL can still bypass DBDock.
Configure the local DBDock stdio MCP with command npx and args [--yes, @dbdock/mcp]. Set DBDOCK_WORKING_DIRECTORY to this linked project's absolute path. Read DATABASE_URL from its local .env file. Use run_sql for changes and ask me to confirm removal of later writes before restoring a checkpoint.
run_sqlSave a checkpoint, run SQL locally and send its operation summary.restore_checkpointRestore locally after confirming that later writes will be removed.sync_recordsRetry pending records without running the change again.~/.claude.jsonclaude mcp add --transport stdio dbdock -- npx --yes @dbdock/mcpSet up the local DBDock MCP in this linked project. Replace DBDOCK_WORKING_DIRECTORY with this project's absolute path. Keep DATABASE_URL in the local .env file. Use run_sql for changes and ask me to confirm before restore_checkpoint. { "mcpServers": { "dbdock": { "command": "npx", "args": [ "--yes", "@dbdock/mcp" ], "env": { "DBDOCK_WORKING_DIRECTORY": "/absolute/path/to/your/project" } } } }
Optional remote MCP for connections stored in Cloud. It uses Cloud credentials and browser sign-in. Use the local setup above to keep your database URL on your machine.