Connect your AI assistant
You can investigate a client's electricity consumption in your AI assistant using the Energiedaten MCP server. This guide covers connection, a first site investigation, and the tools available to your assistant. For an overview, see AI connections.
Before you connect
You need an Energiedaten account, membership in the team you want to connect, and API access for that team. The tools offered to your assistant depend on the team's current access: load-shape analysis and benchmarks require advanced analysis; customer and location tools require the corresponding features.
Your AI client must support remote MCP servers over HTTP and browser-based OAuth authorization. Check whether custom MCP connections are available in your client and account. A client that only runs local MCP servers cannot use this URL directly.
For ChatGPT, you need the desktop app. The direct Energiedaten connection described here is unavailable in ChatGPT on the web and in the mobile apps.
If you are new to Energiedaten, create an account and follow Getting started to connect a meter. An AI connection does not create measurement data. For the first investigation, choose a connected electricity meter with readings for the requested dates. Hourly or finer readings are needed for meaningful power profiles.
Connect your assistant
Choose Claude on the web, ChatGPT and Codex on desktop or another MCP client.
Claude on the web
- In Claude, open Settings, then Connectors. Select Add connector, then Add custom connector.
- Enter a name such as energiedaten.at and the server URL
https://energiedaten.at/mcp. Select Continue. - After the server check, keep authentication set to Always required. Under OAuth client, select No client ID — register one automatically.
- Select Add, then Connect. Sign in to energiedaten.at, choose the intended team and review the access request before authorizing it.
- Return to Claude and open a chat. Ask the connection-check question below. If Claude requests permission to use a tool, review the request and choose Allow once to approve that call.
ChatGPT and Codex on desktop
Open the ChatGPT desktop app and select Settings → MCP servers → Add server. Choose Streamable HTTP, name the server Energiedaten and enter https://energiedaten.at/mcp. Save, select Restart, then Authenticate to sign in and authorize your Energiedaten team. Enter /mcp in the app to see connected servers.
Alternatively, add the server with the Codex CLI:
codex mcp add energiedaten --url https://energiedaten.at/mcp
codex mcp login energiedaten
The desktop app and Codex share the MCP configuration on your computer. See OpenAI’s MCP guide for more details.
Another MCP client
The following is the general setup flow. Labels and connection settings may differ between clients.
-
In your AI client's connection settings, add a remote MCP server. Depending on the client, this setting may be called a connector, integration, or MCP server.
-
Give the connection a name such as Energiedaten and enter this server URL:
Texthttps://energiedaten.at/mcp -
Start the client's OAuth connection flow. Sign in to Energiedaten in the browser window it opens.
-
Select the team whose data the assistant should access. Read the authorization screen, including the information about customer data and processing by your selected AI provider, then authorize the connection.
-
Return to the AI client and enable the connection for your conversation if the client requires it.
Ask your assistant:
Use Energiedaten to tell me which team is connected, then list its metering points and their available data windows.
The assistant can use whoami and list-metering-points to answer. Check the team name before continuing. Switching teams in the Energiedaten app does not switch the AI connection; authorize the connection again to choose another team.
Investigate a site's electricity consumption
Choose a site or metering point from the returned list and dates inside its available data window. Replace the brackets in this example:
Review electricity consumption at [site name] from [start date] to [end date]. Show weekday and weekend patterns, identify the largest demand peaks with their dates and metering points, and highlight gaps affecting the analysis. Include units and the period each result covers.
The assistant can combine consumption totals, load profiles, peak events, and coverage. Ask a follow-up about a particular day or compare two explicit date ranges. Review the returned figures and caveats before using the explanation in a client meeting.
Clients that expose MCP prompt templates can also offer site-deep-dive. This recipe guides the assistant through consumption, base load, day profile, heatmap, peaks, and a benchmark where available. The assistant writes the explanation from tool results; the server does not establish the cause of a peak or produce a formal report.
Compare sites or check data quality
For a portfolio review, ask:
Compare my locations' electricity consumption for [start date] to [end date]. Show feed-in and energy balance where available, and identify incomplete data before ranking the locations.
For a data-quality check, ask:
Check the connected meters for [customer or location] during [start date] to [end date]. Show coverage, the quality-level mix, and the latest delivery for each meter. Identify which gaps limit a comparison.
The portfolio-review and data-quality-check prompt templates guide these workflows in clients that support prompts. You can ask the questions directly in clients that expose only tools.
Tool reference
Your client lists the tools available to the connected team. This catalogue groups the server's tools by purpose.
| Purpose | Tools | Result |
|---|---|---|
| Find the relevant data | whoami, list-customers, list-locations, list-metering-points |
Connected team, customer and location names, metering points, connection status, and queryable data windows. |
| Review consumption | get-consumption-summary, get-consumption-series |
Energy totals and a time series; power figures where the measurement cadence supports them. |
| Understand load shape | get-load-summary, get-day-profile, get-load-heatmap, get-load-duration-curve |
Base load, load factor, weekday/weekend profiles, weekday/hour patterns, and the distribution of load levels. |
| Inspect peaks and compare | get-peak-events, get-benchmark, get-energy-balance |
Timestamped metering-point peaks, a conditional building benchmark, and consumption, feed-in, and balance by location. |
| Assess data quality | get-data-coverage |
Coverage, quality-level counts, and latest delivery information for the selected scope. |
Read the results correctly
Energy is reported in kWh and power in kW. Electricity load profiles require suitable interval data; gas and daily readings do not support the same analyses. Base load is the fifth percentile of the load-duration curve: the level exceeded at least 95% of the time. A peak at one metering point is different from a site's coincident peak across its meters.
Check the returned period, timezone, and coverage. Results use Europe/Vienna time. The effective window may be shorter than requested because of retention, consent coverage, or measurement cadence. One request spans at most 366 days. A series can be aggregated into larger intervals to keep the response bounded.
Missing measurements are returned as null, not zero consumption. A queryable data window does not prove complete data for every day. Quality information distinguishes measured readings (L1), reliable substitute values (L2), and unreliable estimates (L3). A building benchmark may be unavailable when building information or sufficient data is missing.
Troubleshooting
| What you see | What to check |
|---|---|
| The connection does not open an authorization screen | Confirm that the client supports remote HTTP MCP and OAuth, and that you entered the full server URL. Follow your client's instructions for enabling custom connections. |
| Authorization or a query is rejected | Confirm that you still belong to the selected team and that its API access is available. Reconnect if the authorization has expired or been revoked. |
| The assistant shows the wrong team | Authorize the connection again and select the intended team. |
| A tool or prompt is missing | Check the team's access to advanced analysis, locations, or customers. Refresh the client's tool list after changing access. |
| A meter appears but cannot be analysed | Check its connection status and choose dates inside its data window. Follow Connect a meter if it is not connected. |
| Results are empty, partial, or have no power figures | Ask for coverage and the effective date range. Check that readings exist and that their cadence supports the requested analysis. |
| Requests are rate-limited | Wait before retrying and reduce repeated calls. |
Access and revocation
The connection gives the assistant read access to the authorized team's available data, including customer and location information the team manages. Data returned through the connection is processed by your chosen AI provider under that provider's terms. Check those terms and your authority to share customer data before connecting.
To stop further access, open the authorized team in Energiedaten and select Connected apps under Account. Choose the connection and select Revoke. Confirm the action. To use it again, authorize it from the AI client. For chat history or data already received by the provider, use that provider's data-management controls.
