Tooling
You develop against production because local DynamoDB feels optional, then an MCP
agent or a mistyped PartiQL DELETE touches real customer rows. The friction of
bad tooling is not neutral — it pushes teams toward shortcuts that bypass staging.
DynamoDB's console is thin and the CLI is JSON-heavy. A sane workflow uses a local emulator for development, a desktop GUI for browse/edit/export, and guarded agent access when AI drives queries. This section maps that stack: Local and LocalStack, client comparison, data movement, TypeScript codegen, and MCP integrations with read-only clamps when your license is not paid.
What you can do after
- Spin up DynamoDB Local or LocalStack and point DynoTable at it with a dedicated profile.
- Browse, filter, export, and stage edits in a GUI without hand-writing attribute JSON.
- Copy table data across accounts/regions and export to CSV with scoped reads.
- Connect MCP agents with license-aware scopes so read-only mode denies staging.
Reading order
- DynamoDB Local — offline API-compatible dev; what differs from AWS.
- LocalStack — broader AWS emulation when you need more than DynamoDB alone.
- Best GUI clients — what to compare when you pick a desktop tool.
- Books — curated references; not a substitute for access-pattern practice.
- View and edit data — grid workflows, staging, and keyboard-first navigation.
- Copy a table — export/import patterns without a managed copy API in the app.
- Export to CSV — scoped exports and format choices.
- TypeScript types from DynamoDB — codegen from inferred schema samples.
- Query with AI — agent-driven filters and SQL with your own Bedrock models.
- MCP server — agent access surface and license mirroring.
- MCP + Claude Code — wiring for Claude Code sessions.
- MCP + Cursor — wiring for Cursor agents.
- MCP + Codex — wiring for Codex sessions.
- ExtendDB adapter — third-party adapter notes.
DynoTable profiles are single-region. Table prefix filters and "open table by name"
work under resource-scoped IAM when ListTables is denied — a common enterprise
setup. MCP read-only mode clamps to read scope, denies staging, and hides Workbench
and Smart Table tools so an agent cannot commit writes through those surfaces.
Open DynoTable
Download DynoTable and create a Local profile pointed at
http://localhost:8000. Free read-only access (signed-in) covers queries, PartiQL
SELECT, the filter builder, exports, Workbench editor with autocomplete, and saved
queries. Paid trials unlock staging, commits, Smart Table runs, Workbench execution,
and the AI agent.
Stage edits in the item editor and review the diff before commit — the default write path. Connect MCP only after you understand which scopes your license allows; treat PartiQL DML and delete-and-commit shortcuts as direct writes even in paid mode.
Free tools for the same workflow
Several browser tools mirror tasks you would otherwise script against Local or cloud tables:
| Task | Tool |
|---|---|
Marshal fixture JSON before BatchWriteItem | DynamoDB JSON converter |
| Prototype key conditions and update expressions | Expression builder |
| Emit a paginated Query/Scan program | Query builder |
| Estimate Local-vs-cloud cost before deploy | Pricing calculator |
A 2 KB item queried 500 times per second consumes roughly 125 eventually-consistent RCU per second on on-demand billing — Local reports zero, so use the calculators when promoting a Local-tested access pattern to AWS.
The ⌘K palette opens tables, workbench tabs, and AI chats — the fastest path from reading these guides to touching real items.