Which integrations are available for this server?

Provides support for running the MCP server in the Bun JavaScript runtime, which is offered as an optional environment for running the streamstore SDK. Supports running the MCP server in Node.js environments, requiring v20 or greater to run the MCP server through npm. Offers a type-safe TypeScript SDK for interacting with streamstore API, enabling developers to leverage the full capabilities of TypeScript when working with the streamstore platform. Utilizes Zod for runtime type validation, ensuring type safety when interacting with the streamstore API.

How do I use s2-streamstore?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@s2-streamstore show me my recent stream data from the analytics basin" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

s2-streamstore by s2-streamstore

This repo contains the official TypeScript SDK for S2, a serverless data store for streams, built on the service's REST API.

S2 is a managed service that provides unlimited, durable streams.

Streams can be appended to, with all new records added to the tail of the stream. You can read from any portion of a stream – indexing by record sequence number, or timestamp – and follow updates live.

See it in action on the playground.

Quick links:

Runnable examples directory
Patterns package
SDK documentation
S2 REST API documentation

Note: The repository for releases prior to 0.16.x can be found at this link.

Install

npm add @s2-dev/streamstore # or yarn add @s2-dev/streamstore # or bun add @s2-dev/streamstore

Related MCP server: DevDb MCP Server

Quick start

Want to get up and running? Head to the S2 dashboard to sign-up and grab an access key, and create a new "basin" from the UI.

Then define the following environment variables respectively:

export S2_ACCESS_TOKEN="<token>" export S2_BASIN="<basin>"

From there, you can run the following snippet (or any of the other examples).

import { AppendAck, AppendInput, AppendRecord, S2, S2Environment, } from "@s2-dev/streamstore"; const basinName = process.env.S2_BASIN ?? "my-existing-basin"; const streamName = process.env.S2_STREAM ?? "my-new-stream"; const s2 = new S2({ ...S2Environment.parse(), accessToken: process.env.S2_ACCESS_TOKEN ?? "my-access-token", }); // Create a basin (namespace) client for basin-level operations. const basin = s2.basin(basinName); // Make a new stream within the basin, using the default configuration. const streamResponse = await basin.streams.create({ stream: streamName }); console.dir(streamResponse, { depth: null }); // Create a stream client on our new stream. const stream = basin.stream(streamName); // Make a single append call. const append: Promise<AppendAck> = stream.append( // `append` expects an input batch of one or many records. AppendInput.create([ // Records can use a string encoding... AppendRecord.string({ body: "Hello from the docs snippet!", headers: [["content-type", "text/plain"]], }), // ...or contain raw binary data. AppendRecord.bytes({ body: new TextEncoder().encode("Bytes payload"), }), ]), ); // When the promise resolves, the data is fully durable and present on the stream. const ack = await append; console.log( `Appended records ${ack.start.seqNum} through ${ack.end.seqNum} (exclusive).`, ); console.dir(ack, { depth: null }); // Read the two records back as binary. const batch = await stream.read( { start: { from: { seqNum: ack.start.seqNum } }, stop: { limits: { count: 2 } }, }, { as: "bytes" }, ); for (const record of batch.records) { console.dir(record, { depth: null }); console.log("decoded body: %s", new TextDecoder().decode(record.body)); }

Development

Run examples:

export S2_ACCESS_TOKEN="<token>" export S2_BASIN="<basin>" export S2_STREAM="<stream>" # optional per example npx tsx examples/<example>.ts

Run tests:

bun run test

The SDK also ships with a basic browser example, to experiment with using the SDK directly from the web.

bun run --cwd packages/streamstore example:browser

Using S2

S2 SDKs, including this TypeScript one, provide high-level abstractions and conveniences over the core REST API.

Account and basin operations

The account and basin APIs allow for CRUD ops on basins (namespaces of streams), streams, granular access tokens, and more.

Data plane (stream) operations

The core SDK verbs are around appending data to streams, reading data from them.

See the examples and documentation for more details.

Below are some high level notes on how to interact with the data plane.

Appends

The atomic unit of append is an AppendInput, which contains a batch of AppendRecords and some optional additional parameters.

Records contain a body and optional headers. After an append completes, each record will be assigned a sequence number (and a timestamp).

// Append a mixed batch: string + bytes with headers. console.log("Appending two records (string + bytes)."); const mixedAck = await stream.append( AppendInput.create([ AppendRecord.string({ body: "string payload", headers: [ ["record-type", "example"], ["user-id", "123"], ], }), AppendRecord.bytes({ body: new TextEncoder().encode("bytes payload"), headers: [[new Uint8Array([1, 2, 3]), new Uint8Array([4, 5, 6])]], }), ]), ); console.dir(mixedAck, { depth: null });

Append sessions (ordered, stateful appends)

Use an AppendSession when you want higher throughput and ordering guarantees:

It is stateful and enforces that the order you submit batches becomes the order on the stream.
It supports pipelining submissions while still preserving ordering (especially with the s2s transport).

console.log("Opening appendSession with maxInflightBytes=1MiB."); const appendSession = await stream.appendSession({ // This determines the maximum amount of unacknowledged, pending appends, // which can be outstanding at any given time. This is used to apply backpressure. maxInflightBytes: 1024 * 1024, }); const startSeq = mixedAck.end.seqNum; // Submit an append batch. // This returns a promise that resolves into a `BatchSubmitTicket` once the session has // capacity to send it. const append1: BatchSubmitTicket = await appendSession.submit( AppendInput.create([ AppendRecord.string({ body: "session record A" }), AppendRecord.string({ body: "session record B" }), ]), ); const append2: BatchSubmitTicket = await appendSession.submit( AppendInput.create([AppendRecord.string({ body: "session record C" })]), ); // The tickets can be used to wait for the append to become durable (acknowledged by S2). console.dir(await append1.ack(), { depth: null }); console.dir(await append2.ack(), { depth: null }); console.log("Closing append session to flush outstanding batches."); await appendSession.close();

Producer

Streams can support up to 200 appended batches per second (per single stream), but tens of MiB/second.

For throughput, you typically want fewer, but larger batches.

The Producer API simplifies this by connecting an appendSession with an auto-batcher (via BatchTransform), which lingers and accumulates records for a configurable amount of time. This is the recommended path for most high-throughput writers.

Read sessions

Read operations, similarly, can be done via individual read calls, or via a readSession.

Use a session whenever you want:

to read more than a single response batch (responses larger than 1 MiB),
to keep a session open and tail for new data (omit stop criteria).

const readSession = await stream.readSession({ start: { from: { tailOffset: 10 }, clamp: true }, stop: { waitSecs: 10 }, }); for await (const record of readSession) { console.log(record.seqNum, record.body); }

Client configuration

Retries and append retry policy

import { S2, S2Environment, S2Error } from "@s2-dev/streamstore"; const accessToken = process.env.S2_ACCESS_TOKEN; if (!accessToken) { throw new Error("Set S2_ACCESS_TOKEN to configure the SDK."); } const basinName = process.env.S2_BASIN; if (!basinName) { throw new Error("Set S2_BASIN so we know which basin to inspect."); } const streamName = process.env.S2_STREAM ?? "docs/client-config"; // Global retry config applies to every stream/append/read session created via this client. const s2 = new S2({ ...S2Environment.parse(), accessToken, retry: { maxAttempts: 3, minBaseDelayMillis: 100, maxBaseDelayMillis: 500, appendRetryPolicy: "all", requestTimeoutMillis: 5_000, }, }); const basin = s2.basin(basinName); await basin.streams.create({ stream: streamName }).catch((error: unknown) => { if (!(error instanceof S2Error && error.status === 409)) { throw error; } }); const stream = basin.stream(streamName); const tail = await stream.checkTail(); console.log("Tail info:"); console.dir(tail, { depth: null });

appendRetryPolicy: "noSideEffects" only retries appends that are naturally idempotent via matchSeqNum.
appendRetryPolicy: "all" can retry any failure (higher durability, but can duplicate data without idempotency).

Session transports

Sessions can use either:

fetch (HTTP/1.1)
s2s (S2’s streaming protocol over HTTP/2)

You can force a transport per stream:

// Override the automatic transport detection to force the fetch transport. const stream = basin.stream(streamName, { forceTransport: "fetch", });

Patterns

For higher-level, more opinionated building blocks (typed append/read sessions, framing, dedupe helpers), see the patterns package.

Feedback

We use Github Issues to track feature requests and issues with the SDK. If you wish to provide feedback, report a bug or request a feature, feel free to open a Github issue.

Reach out to us

Join our Discord server. We would love to hear from you.

You can also email us at hi@s2.dev.

s2-streamstore