You have devices reporting data to TagoIO around the clock. You have dashboards that show it and an API that serves it. And now you have AI assistants like Claude that are genuinely good at reasoning over data, writing scripts, and answering questions in plain language. The obvious move is to point one at your IoT platform and start asking it things.
But there is a gap. Dashboards and APIs still require you to know where to look and how to query. A dashboard shows what someone decided to put on it. The API answers questions you already know how to phrase. An AI assistant only helps if it can actually reach your live data, understand your device structure, and run queries on your behalf. Without a connection to your real data, it is just guessing.
Therefore you need a wire between the assistant and your IoT data. That wire is the Model Context Protocol (MCP), and the TagoIO MCP Server is the implementation that connects an AI client to your account. This guide walks through setting it up end to end, with Claude Desktop as the example client.
What MCP does, briefly
The Model Context Protocol is an open standard that lets AI assistants talk to external tools and data sources through a consistent interface. Instead of every app inventing its own plugin format, MCP gives the assistant a standard way to discover what a server can do and call it. If you want the full background, read our companion explainer, What Is the Model Context Protocol (MCP) for IoT.
For TagoIO, the MCP server exposes your devices and data so an assistant can read them, answer natural-language questions, and even generate Analysis scripts and dashboards from a description. The assistant does the reasoning. The MCP server does the reaching.
Prerequisites
Before you start, make sure you have these four things:
- A TagoIO account. The free plan covers 5 devices and 5 dashboards, which is plenty for trying this out. Sign up at tago.io if you do not have one.
- A TagoIO token. The MCP server authenticates to your account with a token. Create one scoped to what you actually need (more on scoping later).
- The TagoIO MCP Server. It is open source, with a public GitHub repository and a help-center article.
- An MCP-compatible client. This guide uses Claude Desktop, but the same server works with the Claude apps, ChatGPT, Cursor, Windsurf, and VS Code AI extensions.
Install and configure the TagoIO MCP Server
The TagoIO MCP Server supports two transport modes, and which one you pick depends on where you want it to run.
Local mode (stdio)
In local mode, the MCP server runs as a process on your own machine and communicates with the client over standard input and output (stdio). The client launches the server, the server talks to TagoIO using your token, and everything stays local except the calls to the TagoIO API. This is the simplest setup and a good default for a single developer working on one machine.
You configure local mode by telling your client how to start the server and passing your TagoIO token to it. The exact command and the way you supply the token are documented in the TagoIO MCP help-center article, so follow the current instructions there rather than copying a flag that may have changed between versions.
Remote mode (HTTP)
Version 3.0.0 of the TagoIO MCP Server added remote HTTP support in addition to local stdio. In this mode the server is reachable over HTTP, so you point the client at an endpoint instead of launching a local process. This is useful when you want a shared setup, when the client and server live on different machines, or when you do not want to run a local process at all.
Remote mode shifts where your token lives and how connections are secured, so review the docs for the HTTP configuration and treat the endpoint and token with the same care you would any production credential.
If you only ever need programmatic access from your own code, you do not need MCP at all. The plain TagoIO REST API is still the right tool for that. MCP earns its place when you want natural-language and AI-assisted workflows on top of your data.
Connect Claude Desktop
Claude Desktop reads its MCP servers from a configuration file. To add the TagoIO server, you edit that config file and add an entry that names the server and tells Claude how to reach it: for local mode, the command to launch the server and your token; for remote mode, the HTTP endpoint.
The structure is a JSON object with one entry per MCP server. You give the TagoIO server a name, specify the transport, and provide the credentials. The TagoIO help-center article shows the current config block to paste in, including the exact field names. Use that as your source of truth, then save the file and restart Claude Desktop so it picks up the new server.
The same idea applies to the other clients. ChatGPT, Cursor, Windsurf, and VS Code AI extensions each have their own place to register an MCP server, but the information you provide is the same: a name, a transport mode, and your token or endpoint.
Verify it works
After restarting your client, confirm the connection before you rely on it.
- Check that the client lists the TagoIO server as connected. Most clients show available MCP tools or servers somewhere in their interface.
- Ask a simple question that requires reaching your account, such as listing your devices. If the assistant comes back with your actual device names, the wire is live.
- If something looks off, TagoIO’s Live Inspector is useful for confirming that data is flowing on the device side, so you can tell whether a problem is with the device, the data, or the MCP connection.
Example prompts to run
Once connected, the value shows up in what you can ask. A few prompts to try:
- “List my devices and tell me which ones have not reported data in the last 24 hours.”
- “Show me the latest temperature readings for device X and summarize the trend over the past week.”
- “Write a TagoIO Analysis script that sends an alert when a sensor value crosses a threshold.”
- “Create a dashboard layout that shows the status of all my devices at a glance.”
The assistant uses the MCP server to read your data for the first two, and its own reasoning plus the TagoIO context for the script and dashboard generation. You still review what it produces. Generated Analysis scripts and dashboards are a starting point, not something to ship unread.
Token scoping and security
The MCP server acts with the permissions of the token you give it, so scoping matters.
- Give the token least privilege. If the assistant only needs to read data, do not hand it a token that can delete devices.
- Treat the token as a secret. Do not paste it into shared chats, commit it to a repository, or put it in a screenshot. In remote mode, be especially careful about where the token is stored and who can reach the endpoint.
- Rotate tokens if you suspect exposure, and remove tokens you are no longer using.
The convenience of asking an assistant anything is real, but it operates with whatever access you grant. Grant deliberately.
Troubleshooting common issues
- The client does not show the TagoIO server. Re-check the config file for valid JSON (a missing comma or bracket is the usual cause) and make sure you restarted the client after editing.
- The server connects but cannot read data. This is almost always the token: either it is wrong, expired, or scoped too narrowly for what you asked. Verify the token in your TagoIO account.
- Local mode will not start. Confirm the command and any runtime it depends on are installed and on your path, then check the help-center article for the current launch command.
- Remote mode times out. Confirm the endpoint is reachable from your machine and that nothing in between is blocking the connection.
- Data looks stale or empty. Use Live Inspector to confirm the device is actually sending, so you are not debugging the connection when the real issue is on the device.
The same steps apply elsewhere
Everything above used Claude Desktop, but none of it is specific to Claude. ChatGPT, Cursor, Windsurf, and VS Code AI extensions are all MCP-compatible and connect to the TagoIO MCP Server the same way. Pick the client you already use. The server, the token, and the data are the same on the other side.
Summary
Dashboards and APIs are still the right tools when you know exactly what you want and how to ask for it. MCP fills the gap when you would rather ask in plain language and let an AI assistant do the reaching and the reasoning. To set it up: get a TagoIO account and a scoped token, install the TagoIO MCP Server in local or remote mode, register it in your client’s config, verify the connection with a simple prompt, and keep your token locked down. Then ask your assistant about your devices and let it work against your real data.
Resources
- TagoIO MCP Server documentation
- What Is the Model Context Protocol (MCP) for IoT, the companion explainer
- TagoIO pricing and plans
