> For the complete documentation index, see [llms.txt](https://docs.datacake.de/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.datacake.de/ai-assistant/datacake-mcp-server.md). # Datacake MCP Server Connect your Datacake account to Claude, ChatGPT and other AI assistants via the Datacake MCP server, and query devices, measurements and fleet health in plain language. ## Overview Datacake provides its own **MCP server** (Model Context Protocol). It connects your Datacake account to AI assistants such as **Claude**, **ChatGPT**, **Claude Code** and any other client that supports remote MCP servers. Once connected, the AI can look up your workspaces and devices, read current values and historical measurements, calculate statistics and check fleet health. This lets you go beyond dashboards: instead of only looking at a chart, you can ask what the data means, have anomalies explained and generate complete reports — for example a PDF summary for a factory manager. ### Video {% embed url="" %} ### Details | Setting | Value | | -------------- | ------------------------------------------ | | Server URL | `https://mcp.datacake.co/mcp` | | Transport | Streamable HTTP (remote MCP) | | Authentication | OAuth (sign in with your Datacake account) | | Current scope | Read-only tools (13) | In the first release, the MCP server only offers **read-only** tools. Write actions such as creating devices or dashboards are planned for future releases. ### What is MCP? The Model Context Protocol (MCP) is an open standard that lets AI assistants talk to external software. An MCP server exposes a set of **tools** (for example "list devices" or "get history") that the AI can call on your behalf. With the Datacake MCP server, your AI assistant can: * Answer questions about your workspaces, devices and measurements * Analyse time series data and explain what happened * Create reports, summaries and evaluations from your IoT data * Serve as a building block for your own agents and automations ### Prerequisites Before you start, make sure you have: * A **Datacake account** with access to the workspaces you want to query * An AI client that supports **remote MCP servers with OAuth**, for example Claude (web, desktop or mobile), Claude Code, ChatGPT or another MCP-compatible tool * For Claude: a plan that allows adding custom connectors * For ChatGPT: a plan that supports custom connectors / developer mode {% hint style="danger" %} The MCP server is connected to the **Datacake account you sign in with** during the OAuth step. It has access to **all workspaces and devices** assigned to that account. If you work with several Datacake accounts, sign in with the account you want the AI to use. {% endhint %} ## Setup in Claude You can add Datacake to Claude as a custom connector. A native Datacake connector in Claude's connector directory is planned — once available, setup will be a single click. #### Open the connector settings In Claude, open **Settings** and select **Connectors** in the left-hand menu (under *Customize*). #### Add a custom connector Click **Add** in the top-right corner and choose **Add custom connector**. #### Enter name and server URL * **Name:** `Datacake` * **Remote MCP server URL:** `https://mcp.datacake.co/mcp` Leave the advanced settings at their defaults and click **Continue** / **Add**. #### Connect your Datacake account Datacake now appears in your connector list. Click **Connect**. You are redirected to the Datacake OAuth page, which shows what Claude will be allowed to access: * Full access to your account * Read and change workspaces, devices and data * Stay connected without signing in again Click **Allow access**. You are redirected back to Claude and the connector shows a checkmark. #### Review tool permissions (optional) Click on the **Datacake** connector to see all available tools and decide when Claude may use them. See [Tool permissions](https://claude.ai/chat/08091db7-cbd2-45f3-bc55-de7911e06c0c#tool-permissions) below. #### Enable the connector in a chat Start a new chat, click the **+** icon, open **Connectors** and make sure **Datacake** is switched on. You can now ask questions about your Datacake workspaces. Connectors added in Claude on the web are also available in the Claude desktop and mobile apps. ### Setup in ChatGPT ChatGPT supports remote MCP servers as custom connectors (depending on your plan, this may require enabling **Developer mode**). ### Open connector settings In ChatGPT, open **Settings** and go to **Apps & Connectors** (on some plans called **Connectors**). If custom connectors are not visible, enable **Developer mode** under **Advanced settings**. ### Create a new connector Click **Create** (or **Add custom connector**) and enter: * **Name:** `Datacake` * **Description:** e.g. `Read IoT devices, measurements and fleet health from Datacake` * **MCP Server URL:** `https://mcp.datacake.co/mcp` * **Authentication:** `OAuth` Confirm that you trust this connector and save. ### Sign in with Datacake Complete the Datacake OAuth flow and click **Allow access**. #### Use it in a chat In a new chat, select the Datacake connector from the tools / connectors menu and start asking questions. OpenAI regularly changes menu names and plan availability for custom connectors. If the labels differ from this guide, look for "Connectors", "Apps" or "Developer mode" in the ChatGPT settings. {% endhint %} ## Setup in Claude Code Add the server from the terminal: ```bash claude mcp add --transport http datacake https://mcp.datacake.co/mcp ``` Then start Claude Code, run `/mcp`, select **datacake** and choose **Authenticate**. Your browser opens the Datacake OAuth page. After allowing access, the tools are available in your session. To make the server available in all your projects, add `--scope user`: ```bash claude mcp add --transport http --scope user datacake https://mcp.datacake.co/mcp ``` ## Setup in other MCP clients Any client that supports **remote MCP servers (Streamable HTTP) with OAuth** can connect to Datacake. Most clients use a JSON configuration similar to this: ```json { "mcpServers": { "datacake": { "url": "https://mcp.datacake.co/mcp" } } } ``` The client starts the OAuth flow automatically the first time it connects. For clients that only support local (stdio) MCP servers, use the `mcp-remote` bridge (requires Node.js): ```json { "mcpServers": { "datacake": { "command": "npx", "args": ["-y", "mcp-remote", "https://mcp.datacake.co/mcp"] } } } ``` {% hint style="info" %} The exact configuration file and key names differ between clients. Please refer to your client's documentation on adding remote MCP servers. {% endhint %} ## Authentication and access The Datacake MCP server uses **OAuth**. You never need to create or paste an API token. * **Account binding:** The connection is bound to the Datacake account you sign in with during the OAuth step. * **Scope:** The AI can access all workspaces and devices of that account. * **Session:** You stay connected without signing in again. * **Switching accounts:** Disconnect the connector in your AI client and connect again while signed in with the other Datacake account. * **Revoking access:** Click **Disconnect** on the Datacake connector in your AI client to remove access. {% hint style="warning" %} Treat the connected AI client like any other user with access to your account. For customer or partner workspaces, consider using a dedicated Datacake user that only has access to the workspaces the AI should see. {% endhint %} ## Available tools The current release provides 13 read-only tools: | Tool | Description | | ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------- | | **Who am I** | Verifies the connection and returns the signed-in user. | | **List workspaces** | Lists all workspaces the account can access, with device and member counts. | | **Get workspace** | Details of one workspace: permissions, enabled features and settings. | | **List products** | Lists products (device templates). Every device belongs to a product and shares its fields. | | **List devices** | Paginated, filterable and sortable list of devices in a workspace (by name, tags, product, online status and more). | | **Find device by name or serial** | Finds devices by (part of) their name or by exact serial number / DevEUI across workspaces. | | **Get device** | Everything about one device: identity, online state, tags, metadata and fields. | | **Get current values** | Latest values of selected fields for up to 25 devices in one call. | | **Get history (time series)** | Time series of selected fields for up to 10 devices over a time range. | | **Get field statistics for a time window** | Average, minimum, maximum, sum and change of fields over one or more time windows. | | **Aggregate devices (fleet KPIs)** | Counts and aggregates across all devices matching a filter, computed on the server. | | **Fleet health report** | Health overview of a workspace (optionally by tags): online/offline devices, outages and more. | | **Resolve time range** | Converts presets like *today*, *yesterday*, *last week* or *this month* into exact timestamps. | {% hint style="info" %} Write tools (e.g. creating devices or dashboards) will follow in a later release. This page will be updated accordingly. {% endhint %} ### Tool permissions Most AI clients let you control how each tool may be used. In **Claude**, open **Settings → Connectors → Datacake** to see the list of tools grouped under **Read-only tools**. For each tool (or the whole group) you can choose: | Option | Behaviour | | ------------------ | ------------------------------------ | | **Always allow** | Claude uses the tool without asking. | | **Needs approval** | Claude asks you before every call. | | **Blocked** | Claude cannot use the tool. | Since all current tools are read-only, **Always allow** is a convenient default. If you prefer to see every request before it runs — for example while testing — choose **Needs approval**. In ChatGPT and other clients, similar options are usually offered as approval prompts when a tool is called for the first time. ### Example prompts Once connected, simply describe what you want in plain language. Examples: * *"Which workspaces do I have access to, and how many devices are in each?"* * *"Search for the KEMPER workspace. You will find devices monitoring air quality and noise levels. Create a PDF report for the factory manager."* * *"Which devices have been offline for more than 24 hours?"* * *"Create a fleet health report on online versus offline time for the last month."* * *"Show me the temperature history of device XYZ for last week and explain any unusual peaks."* * *"Compare the average CO₂ values of all rooms for this week and last week."* * *"Find the device with serial number 0004A30B001C1234 and show its current values."* ## Example: AI-generated report A typical workflow for creating a report: 1. **Discovery:** The AI looks up the available Datacake tools and searches for the relevant workspace and devices. 2. **Data retrieval:** It reads current values, historical data and statistics for the relevant fields. 3. **Analysis:** It interprets the data — status by zone, statistics on noise and air quality, short-term episodes and anomalies. 4. **Output:** It creates a report including charts and written explanations of what the data means. Depending on the amount of data, this can take a few minutes. The result can then be refined further, for example: * Add your company logo and corporate design * Focus exclusively on certain devices, zones or measurements * Add a fleet health section with online/offline times * Reuse the report as a template and run it again regularly {% hint style="info" %} This is where the MCP server goes beyond dashboards: a chart shows the data, but the AI explains what happened, why it matters and where you should take a closer look. {% endhint %} ### Troubleshooting **The connector does not appear in my chat** Make sure the Datacake connector is enabled for the current chat (in Claude: **+ → Connectors**). Start a new chat after adding the connector. **OAuth fails or the connector shows "Connect" again** Click **Connect** again and complete the sign-in. Make sure pop-ups are not blocked and that you sign in with the correct Datacake account. **The AI cannot find my workspace or device** Check which account is connected (ask: *"Which Datacake user am I connected as?"*). The AI can only see workspaces and devices assigned to that account. **The AI says it cannot create or change something** The current release is read-only. Creating or modifying devices, dashboards or settings is not yet supported. **Requests for long time ranges are slow or incomplete** Narrow the request to fewer devices, fewer fields or a shorter time range, or ask the AI to use statistics instead of the full time series. **My client only supports local MCP servers** Use the `mcp-remote` bridge as described in [Setup in other MCP clients](https://claude.ai/chat/08091db7-cbd2-45f3-bc55-de7911e06c0c#setup-in-other-mcp-clients). ### FAQ **Does it cost extra?** The MCP server is available to Datacake users. Usage limits and costs of your AI client (Claude, ChatGPT etc.) depend on your plan with that provider. **Is my data used to train AI models?** Data retrieved through the MCP server is passed to your AI client as part of the conversation. How it is processed and stored depends on the provider and your plan settings. Please review the data policy of your AI provider. **Can I restrict access to certain workspaces?** The connection always covers all workspaces of the signed-in account. To limit access, connect with a Datacake user that only has access to the desired workspaces. **Will there be a native connector?** Yes. A native Datacake connector for Claude is planned, so that setup will only require a single click. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.datacake.de/ai-assistant/datacake-mcp-server.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.