kkrpc - TypeScript RPC Library

Build bidirectional RPC systems in TypeScript with full type safety and multiple transport options.

Installation

# npm
npm install kkrpc

# pnpm
pnpm add kkrpc

# Deno
import { RPCChannel } from "jsr:@kunkun/kkrpc"

---

Quick Start

Basic RPC Setup

Server (expose API):

import { NodeIo, RPCChannel } from "kkrpc"

const api = {
	greet: (name: string) => `Hello, ${name}!`,
	add: (a: number, b: number) => a + b,
	counter: 42
}

const rpc = new RPCChannel(new NodeIo(process.stdin, process.stdout), { expose: api })

Client (consume API):

import { spawn } from "child_process"
import { NodeIo, RPCChannel } from "kkrpc"

const worker = spawn("bun", ["server.ts"])
const rpc = new RPCChannel(new NodeIo(worker.stdout, worker.stdin))

const api = rpc.getAPI<typeof api>()

console.log(await api.greet("World")) // "Hello, World!"
console.log(await api.add(5, 3)) // 8
console.log(await api.counter) // 42

---

Core Concepts

RPCChannel

The main class that manages bidirectional communication:

class RPCChannel<LocalAPI extends Record<string, any>, RemoteAPI extends Record<string, any>> {
	constructor(
		io: IoInterface,
		options?: {
			expose?: LocalAPI
			serialization?: { version: "json" | "superjson" }
			validators?: RPCValidators<LocalAPI>
		}
	)

	getAPI(): RemoteAPI // Get proxy to remote API
	expose(api: LocalAPI): void // Expose local API
}

IoInterface (Transports)

Transport adapters for different environments:

| Transport | Class | Environment | | ---------------- | --------------------------------------------------- | ------------------ | | stdio | NodeIo, DenoIo, BunIo | Process-to-process | | WebSocket | WebSocketClientIO, WebSocketServerIO | Network | | HTTP | HTTPClientIO, HTTPServerIO | Web APIs | | Worker | WorkerParentIO, WorkerChildIO | Web Workers | | postMessage | IframeParentIO, IframeChildIO | iframes | | Chrome Extension | ChromePortIO | Chrome extensions | | Electron | ElectronIpcMainIO, ElectronIpcRendererIO | Electron | | Message Queues | RabbitMQIO, KafkaIO, RedisStreamsIO, NatsIO | Distributed |

---

API Definition Patterns

Pattern 1: Inline API (Simple)

const api = {
	greet: (name: string) => `Hello, ${name}!`,
	add: (a: number, b: number) => a + b
}

type API = typeof api

const rpc = new RPCChannel<API, API>(io, { expose: api })
const remote = rpc.getAPI()

Pattern 2: Interface-First (Recommended)

interface MathAPI {
	add(a: number, b: number): Promise<number>
	multiply(a: number, b: number): Promise<number>
}

interface MyAPI {
	math: MathAPI
	greet(name: string): Promise<string>
}

const api: MyAPI = {
	math: {
		add: async (a, b) => a + b,
		multiply: async (a, b) => a * b
	},
	greet: async (name) => `Hello, ${name}!`
}

const rpc = new RPCChannel<MyAPI, MyAPI>(io, { expose: api })

Pattern 3: Nested APIs

interface API {
	math: {
		basic: {
			add(a: number, b: number): Promise<number>
			subtract(a: number, b: number): Promise<number>
		}
		advanced: {
			pow(base: number, exp: number): Promise<number>
			sqrt(n: number): Promise<number>
		}
	}
}

// Usage
const result = await api.math.advanced.pow(2, 10) // 1024

---

Transport Examples

Stdio (Process Communication)

import { spawn } from "child_process"
import { NodeIo, RPCChannel } from "kkrpc"

// Spawn child process
const child = spawn("bun", ["worker.ts"])

// Create channel
const io = new NodeIo(child.stdout, child.stdin)
const rpc = new RPCChannel<LocalAPI, RemoteAPI>(io, { expose: localApi })

// Get remote API
const api = rpc.getAPI()

WebSocket

import { RPCChannel, WebSocketClientIO, WebSocketServerIO } from "kkrpc"

// Server
wss.on("connection", (ws) => {
	const io = new WebSocketServerIO(ws)
	const rpc = new RPCChannel<API, API>(io, { expose: api })
})

// Client
const ws = new WebSocket("ws://localhost:3000")
const io = new WebSocketClientIO({ ws })
const rpc = new RPCChannel<{}, API>(io)
const api = rpc.getAPI()

Web Worker

import { WorkerParentIO, WorkerChildIO, RPCChannel } from "kkrpc/browser"

// Main thread
const worker = new Worker("./worker.ts", { type: "module" })
const io = new WorkerParentIO(worker)
const rpc = new RPCChannel<LocalAPI, RemoteAPI>(io, { expose: localApi })

// Worker thread
const io = new WorkerChildIO()
const rpc = new RPCChannel<RemoteAPI, LocalAPI>(io, { expose: api })

HTTP

import { HTTPClientIO, HTTPServerIO, RPCChannel } from "kkrpc"

// Server
const serverIO = new HTTPServerIO()
const serverRPC = new RPCChannel<API, API>(serverIO, { expose: api })

Bun.serve({
	async fetch(req) {
		if (new URL(req.url).pathname === "/rpc") {
			const response = await serverIO.handleRequest(await req.text())
			return new Response(response, {
				headers: { "Content-Type": "application/json" }
			})
		}
		return new Response("Not found", { status: 404 })
	}
})

// Client
const clientIO = new HTTPClientIO({ url: "http://localhost:3000/rpc" })
const clientRPC = new RPCChannel<{}, API>(clientIO)

---

Advanced Features

Callback Functions

Send functions as arguments that can be invoked remotely:

interface API {
	processData(data: string, onProgress: (percent: number) => void): Promise<string>
}

// Server
const api: API = {
	processData: async (data, onProgress) => {
		for (let i = 0; i <= 100; i += 10) {
			onProgress(i)
			await sleep(100)
		}
		return `Processed: ${data}`
	}
}

// Client
const result = await api.processData("my-data", (progress) => {
	console.log(`Progress: ${progress}%`)
})

Property Access

Access and mutate remote properties:

interface API {
	counter: number
	settings: {
		theme: string
		notifications: { enabled: boolean }
	}
}

// Get values
const count = await api.counter
const theme = await api.settings.theme

// Set values
api.counter = 100
api.settings.theme = "dark"

Enhanced Error Handling

Errors preserve name, message, stack, and custom properties:

class ValidationError extends Error {
	constructor(
		message: string,
		public field: string,
		public code: number
	) {
		super(message)
		this.name = "ValidationError"
	}
}

// Thrown on server
type API = {
	validateUser(data: unknown): Promise<void>
}

// Caught on client
try {
	await api.validateUser({})
} catch (error) {
	console.log(error.name) // "ValidationError"
	console.log(error.message) // "Name is required"
	console.log(error.field) // "name"
	console.log(error.code) // 400
}

Validation (Optional)

Use Standard Schema (Zod, Valibot, ArkType) for runtime validation:

import { RPCChannel, type RPCValidators } from "kkrpc"
import { z } from "zod"

type MathAPI = {
	add(a: number, b: number): Promise<number>
}

const api: MathAPI = {
	add: async (a, b) => a + b
}

const validators: RPCValidators<MathAPI> = {
	add: {
		input: z.tuple([z.number(), z.number()]),
		output: z.number()
	}
}

const rpc = new RPCChannel(io, { expose: api, validators })

Transferable Objects (Browser)

Zero-copy transfer of large binary data:

import { RPCChannel, transfer, WorkerParentIO } from "kkrpc/browser"

interface API {
	processBuffer(buffer: ArrayBuffer): Promise<number>
}

const worker = new Worker("worker.js")
const io = new WorkerParentIO(worker)
const rpc = new RPCChannel<{}, API>(io)
const api = rpc.getAPI()

// Create large buffer
const buffer = new ArrayBuffer(10 * 1024 * 1024)

// Transfer (zero-copy) to worker
const result = await api.processBuffer(transfer(buffer, [buffer]))
// Note: buffer is now detached (length 0)

---

Serialization Options

JSON (Default for Interop)

const rpc = new RPCChannel(io, {
	expose: api,
	serialization: { version: "json" }
})

• Standard JSON
• Works with all interop languages
• No Date, Map, Set, BigInt support

SuperJSON (Default, TypeScript-only)

const rpc = new RPCChannel(io, {
	expose: api,
	serialization: { version: "superjson" }
})

• Supports Date, Map, Set, BigInt, Uint8Array
• TypeScript-to-TypeScript only
• Auto-detected by receiver

---

Common Patterns

Bidirectional Communication

Both sides expose APIs:

// Side A
interface API_A {
	compute(data: number[]): Promise<number>
}

interface API_B {
	notify(message: string): Promise<void>
}

const apiA: API_A = {
	compute: async (data) => data.reduce((a, b) => a + b, 0)
}

const rpc = new RPCChannel<API_A, API_B>(io, { expose: apiA })
const apiB = rpc.getAPI()

// Call B from A
await apiB.notify("Computation complete")

Dynamic API Exposure

Change exposed API at runtime:

const rpc = new RPCChannel(io)

// Later...
rpc.expose(newApi)

Cleanup

Destroy connections when done:

// For transports that support it
io.destroy()

---

Environment-Specific Entry Points

| Environment | Import Path | | ---------------- | ----------------------------------- | | Node.js | kkrpc | | Deno | kkrpc/deno or jsr:@kunkun/kkrpc | | Bun | kkrpc | | Browser | kkrpc/browser | | Chrome Extension | kkrpc/chrome-extension |

// Browser (excludes stdio)
import { RPCChannel, WorkerParentIO } from "kkrpc/browser"

// Deno
import { RPCChannel, DenoIo } from "kkrpc/deno"

// Chrome Extension
import { RPCChannel, ChromePortIO } from "kkrpc/chrome-extension"

---

Error Reference

| Error | Cause | Solution | | -------------------- | ------------------------------ | ------------------------------------ | | RPCValidationError | Input/output validation failed | Check validation schema | | TimeoutError | Request timed out | Increase timeout or check connection | | TransportClosed | Connection closed unexpectedly | Check transport health |

---

Testing

Test with Reference Implementation

// interop/node/server.ts provides a test server
const api = {
	math: { add: (a: number, b: number) => a + b },
	echo: <T>(v: T) => v,
	counter: 42
}

Unit Test Pattern

import { expect, test } from "bun:test"
import { NodeIo, RPCChannel } from "kkrpc"

test("basic RPC call", async () => {
	const api = { add: (a: number, b: number) => a + b }

	// Create connected pair
	const { port1, port2 } = new MessageChannel()

	const serverIO = new NodeIo(port1)
	const clientIO = new NodeIo(port2)

	new RPCChannel<typeof api, {}>(serverIO, { expose: api })
	const client = new RPCChannel<{}, typeof api>(clientIO)

	const result = await client.getAPI().add(2, 3)
	expect(result).toBe(5)
})

---

Best Practices

1. Use Interface-First: Define interfaces before implementations for type safety
2. Handle Errors: Always wrap RPC calls in try-catch
3. Clean Up: Destroy transports when components unmount
4. Validate Inputs: Use validators for public APIs
5. Choose Serialization: Use JSON for interop, superjson for TS-only
6. Transfer Large Data: Use transfer() for ArrayBuffers in browsers
7. Namespace APIs: Use nested objects to organize methods

---

References

• Package: packages/kkrpc/
• Core source: packages/kkrpc/src/
• Serialization: packages/kkrpc/src/serialization.ts
• Adapters: packages/kkrpc/src/adapters/
• Interop guide: skills/interop/SKILL.md