AI-Native eSIM Guide

An eSIM (embedded SIM) is a SIM card that's built into the device — not a physical card you insert and remove. The profile on that SIM (which network it connects to, what data plan it has) is managed remotely via GSMA Remote SIM Provisioning (RSP) protocols.

For an operator, this changes everything. You can provision, modify, and revoke a subscriber's connectivity profile without physically touching their device. A subscriber in Tokyo can switch from a Japanese carrier to a European carrier in seconds — no SIM swap, no retail store.

The catch: managing eSIM fleets at scale requires API access to an Online Charging System (OCS) — the backend that tracks subscriber accounts, balance, package assignment, and network status. Most OCS systems expose a REST API. Most operators access it through a web portal, a custom integration, or both.

Model Context Protocol (MCP) is an open standard — developed by Anthropic, now cross-industry — that lets AI models call external tools in a structured, authenticated way.

Think of it as a type-safe function call interface between an LLM and an external API. Instead of asking Claude to “try to figure out the REST API by reading the docs,” you give it a set of named tools with defined input schemas and structured outputs. Claude calls the right tool, passes the right parameters, and gets back a structured response it can reason over.

MCP is supported natively by Claude Desktop, Claude Code, Cursor, Windsurf, and Smithery. The 2025-03-26 spec adds StreamableHTTP transport — meaning MCP servers can run as hosted services, not just local processes.

eSIM operations are inherently data-rich and repetitive. Every subscriber in your fleet generates a continuous stream of signals: usage patterns, network events, balance changes, package expirations. Acting on that data manually doesn't scale.

AI models are good at exactly this kind of work: reading structured data, finding patterns, and deciding what action to take. But they need a reliable interface to act through — not a browser, not a prompt telling them to “try the API.”

MCP provides that interface. When your AI has Carrier MCP tools, it can:

• Check which subscribers are offline and why
• Identify subscribers at churn risk before they cancel
• Find over-provisioned packages and recommend right-sizing
• Execute bulk package assignments across hundreds of subscribers
• Generate fleet health reports on demand without logging into anything

The key difference from a traditional dashboard: your AI doesn't just display the data — it reasons over it and acts on it. The time between noticing a problem and resolving it collapses from hours to seconds.

Carrier MCP is a Cloudflare Worker that sits between your AI client and the eSIMVault OCS. It exposes 103 MCP tools — covering every OCS operation from subscriber activation to fleet-level analytics.

Authentication uses OAuth 2.1 with PKCE. You authenticate once; Carrier encrypts your eSIMVault API token and stores it in a per-user vault. Every tool call uses your token — there are no shared credentials.

Scope gates enforce what each tier can do. The Free tier gives you all read-only tools — enough to build complete fleet observability. Pro adds write tools (modify subscribers, assign packages). Enterprise adds admin tools (SIM provider-level operations, template management) and white-label options.

Step 1. Install the MCP server in your AI client.

# Claude Code
claude mcp add --transport http carrier https://mcp.carrier.llc/mcp

# Claude Desktop — add to claude_desktop_config.json
{
  "mcpServers": {
    "carrier": {
      "command": "npx",
      "args": ["-y", "@carrierllc/mcp"]
    }
  }
}

Step 2.Authenticate. On first use, Claude will open an OAuth flow. Log in, enter your eSIMVault API token, and close the browser tab. You're done.

Step 3. Run your first report:

Run the fleet_health_report prompt.

You'll get a structured health summary of your entire eSIM fleet — account balances, eSIM status breakdown, low-balance alerts, and utilization rates — in a single response.

Once you have read access working, the natural next steps are:

• Subscriber diagnostics — use diagnose_subscriber to understand why a specific eSIM is offline
• Churn risk scoring — run churn_risk across your highest-value subscribers
• Package optimization — find subscribers on the wrong plan before they notice
• The full MVNO playbook — for a deeper walkthrough of all five core workflows