Discover MCPs & agents
Loading MCPs and agents…
Loading MCPs and agents…
A DynamoDB emulator in Rust, backed by SQLite. Starts in milliseconds as a single static binary, and is verified against real AWS by a public conformance suite. Runs as an HTTP server, an MCP server for coding agents, an embeddable library, or in the browser via WebAssembly.
From the repo.
A DynamoDB emulator in Rust, backed by SQLite. Runs as an HTTP server or an MCP server for coding agents, embeds into Rust and iOS applications as a library, and compiles to WebAssembly for the browser.
The crates.io badge above shows the crate version, which covers the Rust library alone. Everything you install rather than depend on carries a separate product version: the binary, the npm packages, the container images, the browser engine and the Action. That is the number
dynoxide --versionprints and the one to quote in a bug report. Why there are two.
I built Dynoxide because DynamoDB Local is slow, heavy, and can't embed. It needs a JVM, and the typical Docker-based setups adds 2–3 seconds of cold-start, ~205 MB of memory at idle, and a ~225MB Docker image (~473 MB on disk) before you've done anything useful. If you're running integration tests, that's Docker starting, the JVM warming up, and your pipeline waiting.
Dynoxide is a native binary. It starts in milliseconds, idles at ~5.3 MB, and ships as a ~3 MB download. Point any DynamoDB SDK at it and your tests just work.
For Rust projects, there's also an embedded mode - direct API calls via Database::memory() with no HTTP layer at all. Each test gets an isolated in-memory database with zero startup cost. And because it compiles to a native library with no runtime dependencies, it runs on platforms where DynamoDB Local can't, including iOS.
| Metric | Dynoxide (embedded) | Dynoxide (HTTP) | DynamoDB Local |
|---|---|---|---|
| Cold startup | ~0.2ms | ~15ms | ~2,287ms |
| GetItem (p50) | 9µs | 0.1ms | 0.8ms |
| PutItem throughput | ~51,613 ops/s | ~6,703 ops/s | ~945 ops/s |
| 50-test suite (sequential) | ~484ms | ~569ms | ~2,407ms |
| 50-test suite (4x parallel) | ~203ms | ~235ms | ~1,189ms |
Numbers from ubuntu-latest (4-core Intel(R) Xeon(R) 6973P-C, 16GB RAM). Commit 49c6338.
| Metric | Dynoxide (embedded) | Dynoxide (HTTP) | DynamoDB Local | LocalStack (all services) |
|---|---|---|---|---|
| Cold startup | <1ms | ~2ms | ~2,831ms | ~13,473ms |
| GetItem (p50) | 10µs | 0.2ms | 0.7ms | - |
| 50-test CI suite | 573ms | 533ms | 1,871ms | - |
| Full workload (10K items) | - | 2.1s | 8.2s | - |
| Binary / image (download) | ~3 MB | ~3 MB | 225 MB | 1.1 GB |
| Binary / image (on disk) | 7 MB | 7 MB | 473 MB | 1.2 GB |
| Idle memory (RSS) | ~5.3 MB | ~8 MB | ~205 MB | ~388 MB |
The gap is wider on Apple Silicon because the faster CPU amplifies the difference between native code and JVM overhead. Both are real measurements of the same benchmark suite. Full methodology and per-operation breakdowns →
Dynoxide is continuously verified against real DynamoDB by Parity Suite, the DynamoDB conformance suite that runs one test matrix against AWS itself and every major DynamoDB emulator. Pass rates move as the suite grows and each engine changes, so rather than pin a snapshot that goes stale, see the live standings:
Disclosure: Dynoxide and Parity Suite are maintained by the same person. The suite scores Dynoxide on the same public matrix it runs against every other engine, and the results and test code are open.
This covers the native build. The WebAssembly build is scored as its own row: it passes every test it implements, with a far higher skip count than any other target because several operations are still missing.
| Dynoxide | DynamoDB Local | LocalStack (all services) | dynalite | |
|---|---|---|---|---|
| Language | Rust | Java | Python + Java | Node.js |
| Storage | SQLite | SQLite | SQLite (via DDB Local) | LevelDB |
| Runtime dependency | - | JVM | Docker + LocalStack | Node.js |
| Embeddable (Rust / iOS) | ✓ | - | - | - |
| MCP server for agents | ✓ | - | - | - |
LocalStack uses DynamoDB Local internally as its DynamoDB engine, so its startup and memory overhead includes DynamoDB Local's JVM plus LocalStack's own Python routing layer.
Run a local server without installing anything:
npx dynoxide --port 8000
Or install it into a project to pin the version, after which npx dynoxide uses that copy:
npm install --save-dev dynoxide
Or run it in Docker, a drop-in for amazon/dynamodb-local:
docker run --rm -p 8000:8000 ghcr.io/nubo-db/dynoxide
Point any AWS SDK or DynamoDB client at http://localhost:8000. For Homebrew, Cargo, pre-built binaries, and embedding as a Rust library, see the installation guide.
Add --log quiet to suppress informational startup and shutdown messages on
stderr (useful in test suites). Without --log, the output is unchanged.
Dynoxide implements the DynamoDB API across tables, items, query and scan, batches, transactions, PartiQL, streams, TTL, and tags, with GSI and LSI support, the full expression syntax, and DynamoDB-compatible pagination, validation, and error codes. For the operation-by-operation breakdown and a comparison, see the compatibility summary.
Dynoxide is built for local development, testing, and CI, not as a production DynamoDB replacement, so two classes of thing are missing on purpose.
Cloud-only operations with no local equivalent aren't implemented: backups and point-in-time restore, global tables, Kinesis streaming, resource policies, and capacity management. Call one and you get an UnknownOperationException.
A few behavioural differences are also worth knowing when you test against it:
ConsistentRead is accepted but changes nothing. SQLite is strongly consistent, so every read already is - you can't reproduce eventually-consistent reads.DescribeStream returns one shard, and its ExclusiveStartShardId and Limit paging parameters are accepted but ignored.TransactionConflictException, TransactionInProgressException) aren't emulated - there's no concurrent contention in a single process.For the per-feature support matrix, see the live capability matrix; the full operation-by-operation breakdown is in the compatibility summary.
Dynoxide's DynamoDB API semantics and validation logic were informed by dynalite, the excellent DynamoDB emulator built on LevelDB by Michael Hart and now maintained by the Architect team.
Dynoxide is a clean-room Rust implementation. No code was ported directly, but dynalite's thorough approach to matching live DynamoDB behaviour, including edge cases and error messages, was an invaluable reference.
Dynoxide uses SQLite as its storage layer. (AWS's DynamoDB Local also uses SQLite internally.)
Dual-licensed under MIT and Apache 2.0. See LICENSE-MIT and LICENSE-APACHE.
Amazon DynamoDB, DynamoDB, and AWS are trademarks of Amazon.com, Inc. or its affiliates. Dynoxide is an independent project and is not affiliated with, endorsed by, or sponsored by Amazon, and nothing here grants any right to use those names or marks.
Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.