Live Objects - complete developer guide

MCP Protocol (BETA)

Live Objects integrates MCP (Model Context Protocol), a protocol used by LLMs to interact with a platform.

It allows you to use your own LLM based tools to analyze your devices and data on Live Objects. It is very useful for debugging purposes, as it can easily fetch information through several domains (device inventory, data, trigger and action rules, audit logs, access logs…) to detect anomalies and issues on your devices, and to understand the root cause of these issues. It can effectively perform aggregation queries on your data, check your rules configuration, test your decoders, investigate through your audit logs and access logs.

The results you will get is very dependent on the LLM you are using, and the way you ask your questions. So it can be very useful to experiment with different LLMs and different ways of asking your questions to find the best way to get the information you are looking for. Some LLMs are better at understanding complex questions and at using the right tools to fetch specific information.

How many devices do I have in each of my device types?
How many devices have sent data with value.battery.level < 20 in the last 7 days?
Why 'device001' has not sent any data since 2 days?
Who did update the rule 'temperature alert' last time, and what was the change?
What are the most common errors in my audit logs?

Endpoint

MCP is available at the following endpoint:

https://liveobjects.orange-business.com/api/mcp

The authentication is based on the HTTP header X-API-Key. A tool that supports the use of custom HTTP headers is required to use MCP on Live Objects.

You can create a new API Key in the 'Administration > API Keys' section of the Live Objects web portal. Be careful to set up the right permissions for this API key, depending on the information you want to fetch through MCP.

On top of these permissions, you also need to enable the 'MCP' permission on your account. Be careful with the Write permissions, as it allows you to execute actions on Live Objects through MCP, which can potentially be destructive.

Disclaimer

LLMs behaviour can be unpredictable, and can potentially generate destructive actions on the platform if you give them write access.

Be careful about the permissions granted to the API key you configure in your MCP agent, and define only the minimum permissions required for your use case.

Even with read only access, be careful about the information that you allow the LLM to fetch from Live Objects, as it can potentially contain sensitive information about your devices and data. And based on your LLM provider and MCP client, the information fetched from Live Objects can potentially be used by the LLM provider to further train their models, and can be exposed to third parties.

If you use an API Key with write permissions, but have the read only MCP access in your account subscriptions, then the API Key will be restricted to its 'read' permissions. Using an API Key with only 'read' permissions will ensure that the LLM will not be able to perform any destructive action on the platform, even if your account has activated the read/write MCP option.

Enabling MCP on your account

MCP is currently in BETA, and is not enabled by default on all accounts.

If you want to use MCP, please ask for a MCP access through the support page.

There are 2 types of MCP access : read only and read/write.

The read only access allows you to fetch information from Live Objects, but not to perform any action on the platform.

The read/write access allows you to perform actions on the platform (e.g. create or update a device, create a new rule…).

Please ask for the right type of access depending on your needs. But you need to be careful when asking for a read/write access, as it can potentially allow you to perform destructive actions on the platform (e.g. bulk delete devices, delete a rule…). So please make sure to ask for the right type of access depending on your needs.

Configuring your MCP agent

To connect an MCP client (VS Code, Cline, etc.) to Live Objects, you only need to configure:

The MCP endpoint URL: https://liveobjects.orange-business.com/api/mcp
An API Key sent as the HTTP header X-API-Key

For security reasons, do not hard-code your API Key in a file committed to git. Prefer storing it in an environment variable (or your client secret store) and referencing it from your MCP client configuration.

VS Code and GitHub Copilot

If you are using VS Code with GitHub Copilot, you can configure a remote MCP server in your workspace configuration (for example in a {workspace}/.vscode/mcp.json file).

{
  "servers": {
    "liveobjects": {
    "type": "http",
    "url": "https://liveobjects.orange-business.com/api/mcp",
      "headers": {
        "X-API-Key": "${env:LIVEOBJECTS_MCP_API_KEY}"
      }
    }
  }
}

Cline

In Cline, add a new MCP server that points to the Live Objects endpoint and sends the X-API-Key header. Depending on your Cline version, this can be done from the UI settings or by editing its MCP servers configuration file.

{
  "mcpServers": {
    "liveobjects": {
      "type": "streamableHttp",
      "url": "https://liveobjects.orange-business.com/api/mcp",
      "headers": {
        "X-API-Key": "${env:LIVEOBJECTS_MCP_API_KEY}"
      }
    }
  }
}

Once configured, you can validate your setup by asking your agent a simple question such as: "How many devices do I have?". The agent should then be able to query the device inventory through MCP.