hotline

hotline

A messaging channel for Claude Code. hotline is an MCP server that relays a Claude Code session to Telegram, so you talk to your agent the way you text a friend: short bubbles, reactions, tappable buttons, photos.

It speaks Claude Code's experimental two-way channel protocol (claude/channel). That makes it Claude Code specific today. Other harnesses work if they adopt the protocol, or when hotline adds support for them.

The texting experience

Claude replies in bubbles, one thought each, delivered with a typing indicator and a natural pause between them. It reacts with an emoji when a 👍 says it, edits its own messages to show progress, and offers inline buttons when it wants you to pick one thing.

A real exchange looks like this:

you: the build's failing again 😤

claude: ugh again? claude: lemme look claude: yeah it's that flaky integration test, not your code claude: want me to just retry it? [ retry 🔁 ] [ dig deeper ]

You tap a button, the choice lands back in the session as if you had typed it, and the keyboard clears so it can't be answered twice.

Inbound works the same way. You text in bursts, so hotline buffers consecutive messages per chat (1.2s window, 8s max wait) and delivers them as one coherent turn instead of interrupting Claude three times. Photos are downloaded eagerly and handed to Claude as a local path; documents, voice notes, video, and stickers are surfaced as metadata and fetched on demand.

Install

go install github.com/1broseidon/hotline@latest   # -> $(go env GOPATH)/bin/hotline

Requires Go 1.26+. No prebuilt binaries yet.

Quickstart

1. Get a bot token from @BotFather.

2. Drop it in the channel's .env:

   mkdir -p ~/.claude/channels/tele-go
   printf 'TELEGRAM_BOT_TOKEN=123456789:AA…\n' > ~/.claude/channels/tele-go/.env
   chmod 600 ~/.claude/channels/tele-go/.env
   

   A real TELEGRAM_BOT_TOKEN environment variable wins over the .env file.

3. Register hotline as an MCP server in the project you want to text with, via .mcp.json:

   {
     "mcpServers": {
       "hotline": { "command": "hotline", "args": ["run"] }
     }
   }
   

4. Start Claude Code with the channel flag (channels are experimental and loaded by MCP server name):

   claude --dangerously-load-development-channels server:hotline
   

5. DM your bot. The first message from an unknown sender returns a 6-hex pairing code. Approve it from your terminal:

   hotline pair <code>
   

That's it. Your session is now a Telegram chat.

Access model

Nobody talks to your agent without your say-so. The inbound gate decides on the sender, never the chat: a stranger in an allowed group is still dropped.

Policy	Behavior
pairing (default)	Unknown DM sender gets a pairing code; you approve or deny from the terminal
allowlist	Only listed user IDs are delivered; everyone else is silent
disabled	No DMs at all

Groups are opt-in per group ID, with optional requireMention (deliver only when the bot is mentioned, replied to, or a mentionPatterns regex matches) and an optional per-group sender allowlist.

Configuration lives in ~/.claude/channels/tele-go/access.json and is re-read on every inbound message, so edits take effect live:

{
  "dmPolicy": "pairing",          // pairing (default) | allowlist | disabled
  "allowFrom": ["412587349"],     // numeric user IDs
  "groups": {
    "-1001234567890": { "requireMention": true, "allowFrom": [] }
  },
  "mentionPatterns": ["^hey claude"],
  "ackReaction": "👀",            // emoji reaction on receipt ("" disables)
  "replyToMode": "first",         // first | all | off
  "textChunkLimit": 4096,
  "chunkMode": "newline",         // newline | length
  "bubbleMode": "paced"           // paced | instant
}

Pairing codes expire after 24 hours. Approval happens only from your terminal: Claude never approves a pairing or edits access because a chat message asked it to. That request is exactly what a prompt injection looks like.

Outbound is gated too. Every tool call checks the target chat against the same rules before touching the Telegram API, so Claude can only message chats that could message it. The channel also refuses to attach its own state files, and sanitizes uploader-controlled filenames before they enter the message metadata.

The tools