mtcute — TypeScript MTProto Library

What is mtcute?

mtcute is a modern, type-safe TypeScript library for Telegram's MTProto protocol. Unlike Bot API wrappers (grammy, telegraf, node-telegram-bot-api), mtcute speaks Telegram's native binary protocol directly, giving access to the full Telegram API — not just the bot subset.

Architecture overview

Core packages (you likely won't use these directly):

@mtcute/core: MTProto client, high-level methods, TL schema, types
@mtcute/html-parser: HTML ↔ Telegram entities
@mtcute/markdown-parser: Markdown ↔ Telegram entities
@mtcute/file-id: Bot API file_id parsing

Platform packages (use the one that matches your target platform):

@mtcute/node, @mtcute/bun, @mtcute/deno: Runtime bindings for server-side JavaScript
@mtcute/web: Browser bindings (IndexedDB, WebCrypto, WebSocket)

Extra packages (only use these for the specific functionality they cover):

@mtcute/dispatcher: Update routing, filters, bot framework
@mtcute/convert: Importing sessions from other libraries
@mtcute/test: Helpers for unit tests

Always install and import everything from the platform package, not from @mtcute/core:

// ✅ Correct — platform package pre-configures crypto, transport, storage
import { TelegramClient } from '@mtcute/node'

// ❌ Wrong — raw core client requires manual platform setup
import { TelegramClient } from '@mtcute/core'

Basic usage

Creating a client

import { TelegramClient } from '@mtcute/node'

const tg = new TelegramClient({
  apiId: Number(process.env.API_ID),
  apiHash: process.env.API_HASH,
  storage: 'my-account', // SQLite file name (Node.js)
})

Client lifecycle

When you create a TelegramClient, it starts in the disconnected state.

In most cases, you will want to use the start method to connect and authenticate:

// Bot login
const self = await tg.start({ botToken: process.env.BOT_TOKEN })

// User login (interactive)
const self = await tg.start({
  phone: () => input('Phone: '),
  code: () => input('Code: '),
  password: () => input('2FA password: '),
})

However in some cases you may want to manually control the lifecycle:

// Manually connect
await tg.connect()

// Manually disconnect
await tg.disconnect()

// Disconnect and destroy the instance
await tg.destroy()

Calling high-level methods

The TelegramClient class has 300+ high-level convenience methods. Some of the most commonly used ones are listed below:

messages: sendText, sendMedia, getHistory, editMessage, deleteMessages, forwardMessages, searchMessages, ...
chats: getChat, getChatMember, banChatMember, setChatTitle, setChatPhoto, ...
users: getUsers, getMe, setMyProfilePhoto, ...
files: downloadFile, uploadFile, downloadAsStream, ...
dialogs: getDialogs, getPeerDialogs, ...
bots: answerInlineQuery, answerCallbackQuery, setMyCommands, ...
forums: createForumTopic, editForumTopic, ...
stickers: getStickerSet, createStickerSet, ...
stories: sendStory, getStories, ...
auth: start, logOut, sendCode, ...
password: enable2fa, change2fa, ...

// Send a text message
await tg.sendText('me', 'Hello!')

// Send with formatting
await tg.sendText(chatId, html`Hello, <b>${name}</b>!`)

// Send media
await tg.sendMedia(chatId, {
  type: 'photo',
  file: 'file:./photo.jpg', // note the file: prefix!
  caption: 'Check this out',
})

// Get chat info
const chat = await tg.getChat('username')

// Download file
await tg.downloadToFile('path/to/save.jpg', photo)

Some methods have "batch" counterparts that accept arrays instead of single arguments:

const durov = await tg.getUser('durov')
const [durov, telegram] = await tg.getUsers(['durov', 'telegram'])

// Similarly with `getChat` and `getChats`, `getPeer` and `getPeers`

Avoid mixing those up. If the method name is plural, it most likely returns and/or accepts arrays, keep that in mind.

Peer resolution

High-level methods in mtcute accept InputPeerLike for chat/user arguments, which can be:

Numeric ID: 123456789
Username string: 'username'
'me' / 'self' for the current user
A tl.TypeInputPeer object
A high-level object (User, Chat, Message, etc.)

Handling updates with Dispatcher

import { Dispatcher, filters } from '@mtcute/dispatcher'

const dp = Dispatcher.for(tg)

dp.onNewMessage(filters.chat('private'), async (msg) => {
  await msg.replyText('Hello!')
})

dp.onNewMessage(filters.command('start'), async (msg) => {
  await msg.replyText('Welcome!')
})

dp.onCallbackQuery(filters.dataPrefix('btn_'), async (query) => {
  await query.answer({ text: 'Clicked!' })
})

// Start receiving updates
await tg.startUpdatesLoop()

Raw TL method calls

For methods not wrapped by high-level API, use .call():

const result = await tg.call({
  _: 'messages.getEmojiKeywords',
  langCode: 'en',
})
// result is fully typed based on the method's return type

TL type system

mtcute generates TypeScript types from Telegram's TL schema. All types live under the tl namespace:

import { tl } from '@mtcute/core'

TL concept	TypeScript name	Example
Object `user`	`tl.RawUser`	`{ _: 'user', id: number, ... }`
Namespaced object `messages.chats`	`tl.messages.RawChats`
Method `users.getUsers`	`tl.users.RawGetUsersRequest`
Union type	`tl.TypeUser`	`tl.RawUser \| tl.RawUserEmpty`

Every TL object has a _ discriminant field with its constructor name (e.g., _: 'user').

Type guards are available: tl.isAnyPeer(), tl.isAnyMessage(), etc.

The full TL schema is available as JSON at @mtcute/node/tl/api-schema.json (or any platform package). Format:

{
  "l": 223, // layer number
  "e": [ // array of TL entries
    {
      "kind": "class" | "method",
      "name": "user",
      "id": 826896937, // constructor ID
      "type": "User", // return/union type
      "arguments": [{
        "name": "id",
        "type": "int53", // base type name (union or primitive)
        "typeModifiers"?: {
          "predicate"?: "flags.0", // conditional field, e.g. "flags.3"
          "isVector"?: true, // optional: when true, type is actually Vector<type>
        }
      }, ...],
      "comment": "..."
    },
    ...
  ],
  "u": { "User": "..." } // union comments
}

When typeModifiers.predicate is set (e.g. "flags.3"), the field is optional and only present when bit 3 of the flags field is set. When isVector is true, the actual type is Vector<type>.

Looking up types and method signatures

Three bundled tools read from the installed @mtcute/* packages in node_modules. All support fuzzy matching and auto-correct typos.

When to use which:

Need a raw TL type/method definition or its TS interface → get-constructor
Need a high-level client method signature (params, return type, JSDoc) → get-method
Need a high-level wrapper class/interface definition (Message, Chat, etc.) → get-class
Need the TL parameters a raw API call expects → get-constructor --with-references
Not sure which method/class to use → get-method --search or get-class --search

get-constructor: TL schema lookup

Shows TL definition, TypeScript type, union membership, and return type.

node .claude/skills/using-mtcute/tools/get-constructor.js <name>
node .claude/skills/using-mtcute/tools/get-constructor.js --with-references <name>

Accepts any of these input formats:

TL names: user, messages.sendMessage
TS type names: RawUser, TypeInputUser, tl.messages.RawChats, messages.RawGetHistoryRequest

Use --with-references when you need to understand the types referenced by a constructor's arguments or return type — it appends all referenced constructors/unions to the output.

get-method: high-level client method lookup

Shows the full JSDoc + method signature from TelegramClient.

node .claude/skills/using-mtcute/tools/get-method.js <method-name>
node .claude/skills/using-mtcute/tools/get-method.js --search <keyword>
node .claude/skills/using-mtcute/tools/get-method.js --list

Default mode: exact/fuzzy match on method name, falls back to substring in names then descriptions
--search: search in both method names and descriptions, lists all matches
--list: print all available methods with one-line descriptions

get-class: high-level type/class lookup

Shows the full class/interface/type definition with all members from highlevel/types/.

node .claude/skills/using-mtcute/tools/get-class.js <name>
node .claude/skills/using-mtcute/tools/get-class.js --no-members <name>
node .claude/skills/using-mtcute/tools/get-class.js --search <keyword>
node .claude/skills/using-mtcute/tools/get-class.js --list

Default mode: exact/fuzzy match on export name, shows full definition with all members
--no-members: only show the declaration line + JSDoc, not the body
--search: search in both names and descriptions, lists all matches (e.g., --search keyboard)
--list: print all available exports with one-line descriptions

High-level types

mtcute wraps raw TL objects in convenient classes with helper methods:

Message — wraps tl.RawMessage, has .replyText(), .editText(), .delete(), .forward(), etc.
Chat — wraps channel/group/user chat info
User — wraps tl.RawUser
Document, Photo, Video, Audio, Voice, Sticker — media types
InlineQuery, CallbackQuery, ChatJoinRequest — update types

These are in node_modules/@mtcute/core/dist/highlevel/types/.

Text formatting

mtcute supports tagged template literals for formatting:

import { html, md } from '@mtcute/node'

await tg.sendText(chat, html`<b>Bold</b> and <i>italic</i>`)
await tg.sendText(chat, md`**Bold** and *italic*`)

The html and md tags handle escaping automatically when interpolating variables.

Common patterns

Error handling

import { tl } from '@mtcute/node'

try {
  await tg.sendText(chatId, 'hello')
} catch (e) {
  if (tl.RpcError.is(e, 'FLOOD_WAIT_%d')) {
    console.log(`Rate limited, wait ${e.seconds}s`)
  }
}

Sending media

import { InputMedia } from '@mtcute/node'

// Upload from path (note the file: prefix!)
await tg.sendMedia(chatId, InputMedia.document('file:/path/to/file.pdf', {
  fileName: 'report.pdf',
}))

// Upload from buffer/stream
await tg.sendMedia(chatId, InputMedia.photo(buffer))

// Upload from URL
await tg.sendMedia(chatId, InputMedia.photo('https://example.com/image.jpg'))

Keyboards

import { BotKeyboard } from '@mtcute/node'

// Inline keyboard
await tg.sendText(chatId, 'Pick one:', {
  replyMarkup: BotKeyboard.inline([
    [BotKeyboard.callback('Yes', 'btn_yes'), BotKeyboard.callback('No', 'btn_no')],
    [BotKeyboard.url('Docs', 'https://mtcute.dev')],
  ]),
})

// Reply keyboard
await tg.sendText(chatId, 'Choose:', {
  replyMarkup: BotKeyboard.reply([
    [BotKeyboard.text('Option A'), BotKeyboard.text('Option B')],
  ]),
})

Pagination / iteration

Many methods have an iter* counterpart returning AsyncIterableIterator:

// Iterate over chat history
for await (const msg of tg.iterHistory(chatId, { chunkSize: 100 })) {
  console.log(msg.text)
}

// Iterate over search results
for await (const msg of tg.iterSearchMessages({ query: 'hello', chatId })) {
  console.log(msg.id)
}

// Non-iter methods return paginated results with a `.next` cursor
const page = await tg.getHistory(chatId, { limit: 50 })
const nextPage = await tg.getHistory(chatId, { limit: 50, offset: page.next })

Client events

The client exposes Emitter instances for connection lifecycle and errors. Subscribe with .add(), unsubscribe with .remove():

// Connection state: 'offline' | 'connecting' | 'updating' | 'connected'
// 'updating' means connected but catching up on missed updates
tg.onConnectionState.add((state) => {
  console.log('Connection:', state)
})

// Unhandled errors (if no listener, errors are logged to console)
tg.onError.add((err) => {
  console.error('Error:', err)
})

// Raw TL updates (before dispatcher processing)
tg.onRawUpdate.add((update) => { ... })

Proxy / custom transport (Node.js)

import { TelegramClient } from '@mtcute/node'
import {
  HttpProxyTcpTransport,
  SocksProxyTcpTransport,
  MtProxyTcpTransport,
  proxyTransportFromUrl,
} from '@mtcute/node'

const tg = new TelegramClient({
  // or parse from URL (socks4/5, http, https, mtproto):
  transport: proxyTransportFromUrl('socks5://user:pass@1.2.3.4:1080'),
})

using-mtcute