MCP Server Setup for WhatsApp and Instagram Inbox: Cursor, Claude Code, Windsurf (2026)

The Model Context Protocol (MCP) is the new standard for connecting AI agents to external tools. Since Anthropic published the spec in late 2024, every major AI coding tool has added MCP support: Cursor, Claude Code, Claude Desktop, Windsurf, and Zed all speak MCP natively.

InstantReply ships a production MCP server. One config block and your AI agent can read every WhatsApp chat, Instagram DM, and Messenger thread in your inbox, draft replies with your business context loaded, tag hot leads, and close resolved chats. All from inside the tool you already use.

What the InstantReply MCP server exposes

The server ships 12 tools and 6 prompts over stdio transport.

Read tools

• list_conversations: Returns inbox conversations with lead temperature, platform, and last-message preview. Supports filters for open/closed status and date range.
• get_conversation: Fetches one conversation including the most recent messages and contact metadata.
• list_messages: Paginates through the full message history for any conversation.
• list_contacts: Returns your contact list with lead stage and last interaction date.
• get_contact: Fetches one contact record including custom fields, lead stage, and conversation history.
• list_channels: Lists all connected platforms (WhatsApp numbers, Instagram pages, Messenger pages).
• get_analytics_summary: Dashboard stats by date range: reply volume, average response time, AI confidence distribution, open vs closed rate.
• get_usage: Current API usage and headroom against your plan limits.

Write and action tools

• send_message: Delivers an outbound reply to any conversation on any connected channel.
• assign_conversation: Routes a conversation to a specific team member.
• close_conversation: Marks a conversation resolved.
• update_contact: Patches contact name, email, lead stage, or any custom field.

Built-in prompts

• draft_reply: Draft an answer in the team voice, drawing on the last three messages for context.
• summarize_conversation: Three sentences. Customer goal, current state, what unblocks them.
• find_urgent_unanswered: Rank open conversations by churn likelihood.
• weekly_inbox_digest: Wins, regressions, response-time deltas. Paste-ready for standup.
• suggest_macros: Surface repeated answers worth turning into canned replies.
• escalation_note: Internal handoff note. Risk first, context second.

Setup: Cursor MCP server

Cursor's AI panel supports MCP servers natively. Add the InstantReply block to your Cursor config and the AI panel gains all 12 inbox tools in every session.

Config file location: ~/.cursor/mcp_servers.json

{
  "mcpServers": {
    "instantreply": {
      "command": "npx",
      "args": ["-y", "@instantreply/mcp"],
      "env": {
        "INSTANTREPLY_API_KEY": "ir_live_YOUR_KEY_HERE"
      }
    }
  }
}

Replace ir_live_YOUR_KEY_HERE with a key from your InstantReply API settings. Restart Cursor. Open the AI panel and ask: "show me today's hot leads across WhatsApp and Instagram." It will call list_conversations and return ranked results.

Full setup walkthrough: Cursor MCP server for WhatsApp and Instagram.

Setup: Claude Code MCP server

Claude Code runs as a terminal CLI agent. MCP servers are configured in ~/.claude/mcp.json.

{
  "mcpServers": {
    "instantreply": {
      "command": "npx",
      "args": ["-y", "@instantreply/mcp"],
      "env": {
        "INSTANTREPLY_API_KEY": "ir_live_YOUR_KEY_HERE"
      }
    }
  }
}

After adding this block and restarting Claude Code, the CLI agent can pull inbox data directly: ask it to "show today's open Instagram DMs sorted by lead temperature" and it will call list_conversations with the right filters, then format the output. You can ask it to draft a reply for any specific chat and it will call get_conversation to read history before drafting.

Full walkthrough: Claude Code MCP server for WhatsApp and Instagram.

Setup: Claude Desktop MCP server

Claude Desktop adds MCP servers via the application config file.

Config file location (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "instantreply": {
      "command": "npx",
      "args": ["-y", "@instantreply/mcp"],
      "env": {
        "INSTANTREPLY_API_KEY": "ir_live_YOUR_KEY_HERE"
      }
    }
  }
}

After adding the block and restarting the app, ask Claude: "give me a morning triage of all open DMs." Claude calls list_conversations, sorts by lead temperature, and formats a clean report. You can then ask it to draft replies for the top three leads in one follow-up prompt.

Full walkthrough: Claude Desktop MCP server for WhatsApp and Instagram.

Setup: Windsurf MCP server

Windsurf's Cascade agent supports MCP natively. Config goes in the Windsurf MCP config file.

Config file location: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "instantreply": {
      "command": "npx",
      "args": ["-y", "@instantreply/mcp"],
      "env": {
        "INSTANTREPLY_API_KEY": "ir_live_YOUR_KEY_HERE"
      }
    }
  }
}

Cascade can now run multi-step inbox flows: fetch hot leads, draft replies using your AI brain context, present all drafts for one-click approval, and send on confirm. Full walkthrough: Windsurf MCP server for WhatsApp and Instagram.

Setup: Zed MCP context server

Zed calls MCP servers "context servers." The config uses a slightly different shape in Zed settings.

Go to: Zed menu > Settings > Assistant > Context Servers

{
  "context_servers": {
    "instantreply": {
      "command": {
        "path": "npx",
        "args": ["-y", "@instantreply/mcp"]
      },
      "settings": {
        "INSTANTREPLY_API_KEY": "ir_live_YOUR_KEY_HERE"
      }
    }
  }
}

The Zed assistant panel can then pull inbox context directly. Ask it to "show me today's Instagram DMs from hot leads" and it will surface the list inline without switching apps. Full walkthrough: Zed MCP server for WhatsApp and Instagram.

What you can do once any agent is connected

The 12 tools cover the full inbox management loop. Here are the most common tasks teams run via MCP agents:

Morning triage in one prompt

Ask any agent: "show me all open conversations from the last 24 hours sorted by lead temperature, very hot first." The agent calls list_conversations with the right parameters and returns a prioritized list. You pick which ones to reply to. No tab switching. No manual scrolling.

Draft a reply using conversation history

Ask: "draft a reply to conversation ID 12345 using our standard pricing tone." The agent calls get_conversation to load history, then drafts a reply that matches the conversation flow. You review and ask it to send_message once it looks right.

Tag and qualify leads in bulk

Ask: "find all conversations from the last week where the contact asked about pricing and tag them as hot leads." The agent calls list_conversations, filters by message content, then calls update_contact for each match. A task that would take 20 minutes of manual clicking runs in under 30 seconds.

Weekly inbox digest for standup

Use the built-in weekly_inbox_digest prompt. It calls get_analytics_summary for the past 7 days and formats wins, regressions, and response-time changes in a paste-ready block for your team standup.

The niche: AI inbox management via MCP

Most SaaS tools for WhatsApp and Instagram have no MCP server. The workflow is always the same: open the dashboard, click the conversation, type or paste a reply. That works at low volume. At scale, or for developer and founder teams who already live inside AI tools like Cursor and Claude Code, that tab-switching is constant friction.

The InstantReply MCP server removes that friction entirely. Your AI agent already understands context, tone, and intent from whatever task you are working on. It can use that context to write better inbox replies than you would get from a standalone dashboard. And the 12 tools cover everything from reading history to closing the chat, so the agent can complete the full loop without handing off to the browser.

Free on every plan. No credit cap. No per-call charge. See the full MCP server documentation or start a 10-day Pro trial to test it with a real key.