Control Claude on your laptop via text messages

TextMe is a skill for setting up the njerschow/textme daemon, which bridges inbound iMessages through Sendblue to a Claude Code session on the user's machine. A whitelisted phone can send text, images, or voice notes and request filesystem work, code execution, or directory changes. It turns the phone into a remote interface for a session running on an online workstation.

Built-in commands show status, queued messages, history, and command help. Other commands return to the home directory, reset conversation history, change directories, or stop the current task. The yes and no reply paths approve or reject pending actions in the underlying session. Other messages become prompts for the active conversation.

Use it when the user wants to initiate, supervise, or interrupt local work from a phone while away from the keyboard. It can also support a long-running headless workstation. It is an inbound control workflow, while sendblue-notify handles outbound completion alerts. Outbound-only needs do not require this daemon.

The host must stay online and run macOS or Linux with Node.js 18 or later, an authenticated Claude Code installation, and an active Sendblue account with a provisioned number. Voice-note transcription needs a separate OpenAI API key. Without that optional setup, voice notes may be dropped or remain untranscribed depending on the daemon version.

This is one conversation on one machine rather than a multi-tenant service. Two daemons sharing a Sendblue number would compete to process the same inbound messages. Delivery is polled instead of pushed, so interaction includes polling latency.

Setup inputs include Sendblue API credentials, its phone number, the user's controlled phone number, polling interval, and conversation-window size. The daemon's API credentials differ from the CLI's local bearer token. Messages and media from allowed senders become session input, and replies return through Sendblue. Start the daemon in the foreground with:

npm start

Verification precedes persistent operation. The workflow checks a status response, a benign command result, rejection of a non-whitelisted sender, and complete cessation of processing when the daemon is stopped. If any check fails, automatic startup is not enabled. Initial checks should use a sandbox directory and a regular user account rather than root.

The phone-number whitelist is the sole authorization boundary between an inbound message and host execution. Start with one controlled number, remove obsolete entries, and account for number spoofing or hijacking. The daemon inherits its user's ability to read, write, and execute, so a dedicated host or VM can limit exposure. Shared machines and directories containing sensitive files need deliberate consideration.

Configuration contains credentials and personal identifiers. It stays outside repositories, backups, cloud sync, shared logs, and screenshots, with restrictive file permissions. Messages remain untrusted input and must not bypass the session's review through eval or shell substitution. Replies can appear on a lock screen, so secrets and customer data should not be requested over text.

TextMe uses Sendblue for inbound polling and outbound replies, Claude Code for the local session, and optional OpenAI Whisper for voice transcription. pm2 or macOS launchd can maintain the daemon after verification. Polling should remain at least five seconds unless rate and cost implications are understood.

Sendblue contact verification also affects outbound replies. sendblue-cli handles account setup and credential management, sendblue-api supports custom handlers, and update-config covers related hook edits. A Sendblue MCP alternative exposes outbound tools without opening this inbound control channel. sendblue-notify can complement TextMe for bidirectional work.

It serves users who want to control their own online workstation from a phone. It is useful when remote initiation and supervision are needed, provided the whitelist and host privileges match the intended trust level.