InduQT for Claude: connect MQTT, Node-RED, OPC UA and Excel data, find the root cause

Add the InduQT connector to Claude and troubleshoot industrial equipment from its real signals. Connect an OPC UA server, an MQTT broker, Node-RED, a historian export or any proprietary feed from the conversation, add manuals and photos, and let InduQT’s decision engine find the anomaly and its root cause.

Add InduQT to Claude See live feeds

Add InduQT to Claude

Open InduQT in Claude’s directory. Open https://claude.ai/directory/connectors/induqt (or search “InduQT” under Settings → Connectors → Browse connectors in claude.ai or Claude Desktop) and click Connect.
Sign in and allow. Claude opens an InduQT page: sign in (the first sign-in creates a free account) and click Allow. Claude can then read your machines and start investigations, but cannot delete anything.
Ask. Start with “What can InduQT do for me?” or one of the prompts below. New accounts can try the public live demo machines straight away.

InduQT is listed in Claude’s connector directory. Other ways in: a custom connector with the URL https://api.induqt.ai/mcp, or in Claude Code claude mcp add --transport http induqt https://api.induqt.ai/mcp and then /mcp to sign in. Details for every client: MCP docs.

What to ask

OPC UA

“Browse the OPC UA server at opc.tcp://gateway.example.com:4840, find the boiler’s drum level, feedwater flow and steam pressure, and connect them to InduQT.”

MQTT

“Preview plant/+/telemetry on mqtts://broker.example.com:8883 and connect the compressors, one machine per topic.”

Node-RED or a proprietary feed

“Our Node-RED box reads the press over Modbus. Give me an InduQT ingest key and a flow to import.”

Excel or CSV

“Load last week’s historian export from this Google Sheets link into Boiler B-101.”

Manuals and photos

“Add the burner manual from this link to B-101. Then tell me what lockout code 21 means for this machine.”

Root cause

“B-101 tripped on low water at 06:10. Which signal moved first, what is the likely cause and what should I check?”

Tested on public data

Everything below was done through the connector’s tools, the way Claude calls them, on 27 October 2026.

Connecting live sources

OPC UA: browse_opcua_server walked the public sample server (opcua.demo-this.com) from the Objects folder down to Boiler #1’s drum level node; create_connector mapped drum level, feedwater and steam flow; live values arrived within 40 seconds.
MQTT: preview_mqtt on Helsinki’s open transport broker showed 2.4 messages a second for one tram with its numeric fields; the connector now streams four trams’ speed, acceleration, delay and doors.
Proprietary feed: create_ingest_key returned a site key and a Node-RED flow; a script posting 40 minutes of press data was accepted (160 readings). With a hydraulic pressure drop seeded at minute 28, the decision engine answered leak or demand above capacity and named hydraulic pressure as the first mover, at the right minute, with motor current a minute later.

Root cause on a labelled fault data set

We took four seeded faults from the U.S. Department of Energy’s LBNL Fault Detection and Diagnostics boiler-plant data (a simulated two-boiler hot-water plant) and built 2½-hour windows that run fault-free until 07:30 and faulted after it. Each was loaded by link with import_file_url, evaluated with evaluate_machine and investigated with start_investigation, given only “not behaving normally since about 07:30”.

Seeded fault	Decision engine	AI investigation (top cause)
Return-water temperature sensor bias	Sensor fault ✔	Return temperature sensor bias, 80% ✔
Boiler supply temperature sensor bias	Sensor fault ✔	Supply temperature sensor fault, 80% ✔
Boiler temperature loop badly tuned (PI)	Control loop hunting ✔	Demand change ✘ (control hunting second, 55%)
Boiler fouling to 95% effectiveness	Missed ✘	Missed ✘

Three of four for the decision engine, two of four first-choice (three of four in the top two) for the AI investigation. Fouling raises fuel use by a few percent over weeks; it does not show in a two-hour window and needs a long-term trend, which is what InduQT’s check for it asks for. These windows were also used to improve the engine (oscillation detection on large temperature values, and background changes before the reported start), so treat them as a demonstration rather than an independent benchmark. The windows are public: README and CSVs.

Tools Claude gets

37 tools, each with a title and read-only or write hints. The ones that matter most:

get_started: Call first: what the account has and the next step for this user.
search_troubleshooting_knowledge: Fault codes, causes and checks for boilers, drives, CNC, servo, compressors, pumps and chillers, with sources. Free.
diagnose: AI diagnosis of a symptom or alarm, optionally with a machine’s live data.
list_demo_machines: Public demo machines streaming live (OPC UA, MQTT, CSV link).
get_demo_data: Time series for a public demo machine.
browse_opcua_server: Browse an OPC UA server’s nodes and current values before connecting it.
preview_mqtt: Listen to an MQTT broker for a few seconds: topics, rates, sample payloads, numeric fields.
create_ingest_key: A site-scoped ingest key with curl and Node-RED examples, for gateways, historians and proprietary feeds.
add_documents_from_links: Fetch manuals, photos and work-order exports from links into a machine’s documents.
get_upload_link: The page where the user adds local photos, trend exports and PDFs.

Full list: MCP docs.

Questions

What is the InduQT connector?

A remote MCP (Model Context Protocol) server at https://api.induqt.ai/mcp. Claude calls its tools to read your machines’ signals, connect new data sources, search manuals and work orders, and run InduQT’s decision engine and AI diagnosis.

Which data sources can I connect?

OPC UA servers and MQTT brokers reachable from the internet, Azure IoT Hub / Event Hubs, CSV, Excel and JSON files or links (Google Sheets, Drive, Dropbox), and anything that can make an HTTPS request — Node-RED, PLC gateways, historians and proprietary systems — through a site-scoped ingest key. Servers inside the plant network connect through Node-RED or a small script on site.

Can I upload PDFs and photos from Claude?

Files attached in the chat do not reach connectors, so InduQT gives Claude two ways: add_documents_from_links fetches public links (a maker’s manual, a shared Drive file), and get_upload_link gives you the page to drop local photos, trend exports and PDFs. InduQT reads them, scanned pages and alarm-screen photos included, and cites the page in the diagnosis.

How does InduQT find the root cause?

Rules first: every signal is profiled for its baseline, when it started to move, in which direction and whether it oscillates, and the equipment family’s causes are scored against that evidence; changes that were already under way before the reported start count as background. A fast model chooses only among the rules’ candidates and safety rules have the last word. The AI diagnosis then reads the data, the engine’s verdict, the machine’s manuals and work orders and your team’s confirmed fixes.

Is my data safe?

Claude signs in with OAuth and gets read and write access to your account only; it cannot delete sites, machines or data. Access tokens last an hour and refresh tokens rotate. You can disconnect Claude any time in InduQT Settings.

What does it cost?

The connector is free to add. Reading data, the knowledge base and the public demo are free; AI diagnoses count toward your InduQT plan, and Claude says so before it runs one.

LBNL FDD Data Sets: Lawrence Berkeley National Laboratory and Pacific Northwest National Laboratory, U.S. DOE Building Technologies Office. DOI 10.25984/1881324.