Send iMessages from Claude

What does your team get when Claude can operate the line?#

The gap was never writing. It's pulling lists, checking reachability, personalizing openers, and pasting them one at a time. With the MCP server, point Claude at a CSV and it handles availability checks, drafting, and status tracking.

• Blue bubbles with read receipts and typing indicators use the same thread colleagues and family do. This looks like a person, not a platform, which is why customers book 2-3x more appointments where a booking funnel exists.
• Your line is dedicated, unlimited iMessages, with automatic RCS then SMS fallback. Lines pace at 50 opted-in / 30 cold new contacts a day; replies never metered. No A2P registration required, no brand vetting sits between your team and the first send.

How do you connect Claude to Blue Reacher?#

The server is an npm package that runs over stdio in Claude Code, Claude Desktop, and Cursor. You need Node 18+ and an API key from your dashboard; keep it in MCP config or secrets manager, never in a prompt or repo.

• If nobody wants to touch a config file, white-glove setup covers it: dedicated line, message intent tuning, and MCP server wiring for your chosen client. Full schemas, shapes, rate limits, and error codes at docs.bluereacher.com.

• Claude Code, one command: claude mcp add bluereacher --env BLUEREACHER_API_KEY=brk_live_your_key -- npx -y @bluereacher/mcp
• Confirm it connected by running claude mcp list, which shows the server and its tool count
• Claude Desktop, paste into claude_desktop_config.json (Cursor uses .cursor/mcp.json): {"mcpServers": {"bluereacher": {"command": "npx", "args": ["-y", "@bluereacher/mcp"], "env": {"BLUEREACHER_API_KEY": "brk_live_your_key"}}}}
• Restart the client after saving. The tools appear under a bluereacher group
• Rotating the key in your dashboard takes effect on the next tool call, and a bad key returns a clear 401 instead of failing quietly

What does an operator actually type?#

The prompts below are what a sales manager types, not code. The pattern that holds up in practice: ask for the read first, approve what comes back, then ask for the send.

• "Check whether +1 415 555 0198 can receive iMessage. Report the answer, do not send anything."
• "Read trials.csv, check iMessage availability for all 12 numbers, then draft one message per person referencing their company. Show me the drafts. Do not send."
• "Send the three drafts I just approved from line_9f2c1a, one at a time, and report each status."
• "Pull the last 50 messages on that line and summarize every reply that mentions pricing."
• "External id crm-lead-4471 went out this morning. Did it deliver?"
• "Show me every message from the last hour with status failed, with the recipient number for each."
• "List my active lines and tell me which one is labeled for West Coast."

Where does a human have to stay in the loop?#

One of the five tools touches a real person. Treat send_message differently from the other four: reading your pipeline is convenient, but sending unattended is a liability with your company's name on it.

• Consent, quiet hours, and pacing still apply; STOP opt-outs auto-detected. No A2P registration doesn't mean permission to text strangers; point Claude at defensible lists and pace like a person would.

• Run read-only by default. Set BLUEREACHER_MCP_READ_ONLY=true to expose only availability checks, message lists, status, and lines
• Scope the key. Issue one API key per agent, restricted to the line_id that agent may use, so a bad prompt can't reach your other lines
• Require approval on sends. In Claude Code, keep send_message off the auto-approve list so every call surfaces a confirmation
• Keep a human on bulk work. Have Claude draft and show the full batch, then send only after you approve it. One message per call, no unattended loops
• Log the returned message id on every send so each message ties back to the record that triggered it
• Watch the activity feed. Every tool call lands in your dashboard with the key that made it