MCP server for IoT SIMs
Updated
The M2M.global MCP server lets an AI agent manage your IoT SIMs the way an engineer would: list them, read their data usage, activate or suspend them, top them up. It speaks the open Model Context Protocol, so it works with Claude, ChatGPT, Cursor and any MCP-compatible client. What the agent can do is set by the token you give it, tool by tool.
Status: Coming soon (checked 26 October 2026) · Notify me when available
Key points
- Read and act. The agent does not only report: with a write token it can change the state of a SIM.
- One tool, one API endpoint. The MCP server is generated from the public API, so permissions, limits and logs are the same whether a human script or an agent makes the call.
- Permissions per token. An agent only sees the tools its token allows. A tool it cannot see, it cannot call.
- Writes are labelled. Every tool that changes real data says so first in its description (WRITE), so the agent and its user know before the call.
What an agent can do
| Task | Example request to the agent | Changes data? |
|---|---|---|
| List SIMs and lines | “Which SIMs have not sent data for 7 days?” | No |
| Read data usage | “Show the top 10 SIMs by data used this month.” | No |
| Read events | “What happened to the SIM in truck 42 yesterday?” | No |
| Suspend or reactivate a line | “Suspend the SIM of the stolen tracker.” | Yes |
| Top up a line or a data pool | “Add 1 GB to the pool used by the cameras.” | Yes |
| Manage data pools | “Move these 20 SIMs into the warehouse pool.” | Yes |
| Follow orders | “Has my last order of SIM cards shipped?” | No |
Ordering new SIMs from an agent will follow the same rules once ordering is open in the API.
Connect your agent
The server is reached over HTTPS with a bearer token created in your account. Example configuration for an MCP client that supports remote servers:
{
"mcpServers": {
"m2m-global": {
"type": "http",
"url": "https://mcp.m2m.global/",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
The address above is the planned one: it answers once the server opens (see the status at the top of this page). Use a read-only token first, check what the agent does, then give it a write token if you want it to act.
Safety rules
- Start read-only. Give a new agent a read-only token; it will not even see the tools that change data.
- Read before write. The server asks agents to read the current state before changing it.
- Explained refusals. Errors carry a reason (
application/problem+json), so an agent stops instead of repeating a forbidden call. - Same rules as the API. An agent has exactly the rights of its token, no more, through MCP or the API.
Why it matters
IoT fleets are run by small teams. An agent that can answer “which devices are silent?” and then act on the answer, inside the limits you set, saves the repetitive part of the job. Combined with private delivery, the agent manages the SIMs while the device data itself stays on private links, out of the public internet.
Learn more
Frequently asked questions
What is an MCP server?
MCP (Model Context Protocol) is an open protocol that lets AI assistants call tools exposed by a service. An MCP server for IoT SIMs lets an assistant such as Claude, ChatGPT or Cursor list your SIMs, read their data usage and, if you allow it, change their state.
Can the AI agent change my SIMs, or only read them?
Both, depending on the token you give it. A read-only token exposes only reading tools. A write token also exposes tools that activate, suspend or top up SIMs; their descriptions start with WRITE so the agent and you know which calls change real data.
What happens if the agent makes a wrong request?
Every refusal comes back as a structured error (application/problem+json) whose detail field says why. The agent can read it and correct itself instead of retrying blindly.
Is the MCP server different from the API?
No. Each MCP tool maps to one endpoint of the public API, with the same permissions. Anything an agent does, your own code can do through the API, and the reverse.