LabJack T-series DAQ — MCP Server

Let an AI agent read analog inputs, stream waveforms, read thermocouples, check digital lines and (with your permission) set DACs and digital outputs on LabJack T4, T7/T7-Pro and T8 DAQ devices through LabJack's official LJM library.

Try it without hardware

uvx labmcp-labjack --simulate --check
uvx labmcp-labjack --simulate --option sim_model=T8 --check     # T4, T7, T7-Pro (default) or T8

The simulator wires a 2 V / 10 Hz sine to AIN0, a thermistor divider (~2.5 V) to AIN1, DAC0 back into AIN2, an LM34 cold-junction sensor to AIN3, and a ~37 °C thermocouple to AIN0-AIN3 when the thermocouple feature is used. Other inputs are grounded.

Connect your LabJack

1. Install LJM (the native library and USB driver) from labjack.com/support/software/installers/ljm. The labjack-ljm Python package is installed automatically with this server, but it needs that native library.
2. Close other LabJack software (Kipling, LJLogM, LJStreamM). A device can only be claimed by one process at a time over USB.
3. Test the connection:

   uvx labmcp-labjack --check                                  # first LabJack found
   uvx labmcp-labjack --address 470012345 --check              # by serial number
   uvx labmcp-labjack --address 192.168.1.207 --option connection_type=ETHERNET --check
   uvx labmcp-labjack --option device_type=T4 --check          # only open a T4

   --address is the LJM identifier: a serial number, an IP address, a device name, or ANY (the default).

Add to your MCP client

Claude Code

claude mcp add labjack -- uvx labmcp-labjack --address 470012345

Claude Desktop / Cursor / Windsurf (claude_desktop_config.json, .cursor/mcp.json, …)

{
  "mcpServers": {
    "labjack": {
      "command": "uvx",
      "args": ["labmcp-labjack", "--address", "470012345"]
    }
  }
}

Add --read-only to allow measurements but block the DAC and digital-output tools (set_outputs_safe stays available). Add --option safe_dio=FIO0,FIO1 to make set_outputs_safe always release those lines, even if another program drove them. For other clients, generate the snippet with uvx labmcp config labjack --client vscode (also cursor, codex, claude-code).

Tools

get_connection_info, get_command_log and reconnect are built into every LabMCP server. The command log records every LJM register read and write.

Safety limits

Override at launch, e.g. --limit max_dac_voltage_v=2.5. Limits are checked before anything is sent. Streams are also capped at 2 million samples in total.

Example prompts

• "Read AIN0-AIN3 on the ±1 V range and tell me which channels look like they are floating."
• "Stream AIN0 and AIN1 at 5 kHz for 2 seconds, give me the RMS and peak-to-peak of each, and save the data to ~/data/run12.csv."
• "Read the type K thermocouple on AIN2 every 10 seconds for a minute and tell me whether it is still warming up."
• "Set DAC0 to 1.25 V, then check the loopback on AIN2 and report the error in millivolts."
• "Which digital lines are outputs right now, and what level are they at?"
• "We're done — put all the outputs back in a safe state."

Notes

• Model differences are enforced. Valid input ranges (AIN#_RANGE) and resolution indices come from LabJack's register map: T7 ±10/±1/±0.1/±0.01 V and index 0-8 (T7-Pro 0-12, but 9-12 cannot stream); T8 ±11 … ±0.018 V and index 0-16 (one index shared by all inputs); T4 fixed ranges (AIN0-3 ±10 V, AIN4-11 0-2.5 V) and index 0-5. DAC range: T4/T7 0-5 V, T8 0-10 V.
• Settings persist. Range, resolution and differential settings made by read_analog_inputs stay on the device until changed or power-cycled. Omitted settings are left as they are; every reading reports what it was measured against (GND, the T7 negative input, or the T8's isolated terminal).
• Reading digital lines never changes their direction. read_digital_inputs uses the DIO_STATE/DIO_DIRECTION bitmasks. (Reading a single-line register such as FIO3 would switch that line to input.)
• T4 flexible I/O. AIN4-AIN11 share terminals with FIO4-EIO3. The server refuses to read those AINs while the line is a digital output, because the read would switch it to analog and release whatever it drives.
• set_outputs_safe default is input, the T-series power-up state (input with pull-up). If an attached relay board treats a pulled-up input as "on", use dio_mode="low" instead.
• Thermocouples use the AIN extended feature (AIN#_EF_INDEX 20-30). With cjc="internal" the CJC register is set to its documented default TEMPERATURE_DEVICE_K (address 60052), which the T8 firmware maps to each channel's own terminal sensor (TEMPERATURE#_CAPTURE); cjc="lm34" uses LabJack's LM34 conversion (55.56 K/V, 255.37 K). An open input returns valid=false (the device reports -9999).
• Streams follow LabJack's stream_basic.py example (internal clock, no trigger, auto settling). While a stream runs, command-response analog reads are blocked by the device. Skipped scans (buffer overflow) are reported and left empty. set_outputs_safe does not wait for a running stream: it stops it at the next read (the partial result is returned with aborted: true). A stream runs at 1 scan/s or faster (so that stop is prompt) and always takes at least two scans; that real duration is what the max_stream_duration_s limit checks. The returned waveform is downsampled by keeping each block's minimum and maximum, so short spikes are not lost; save_path must be a new .csv file (an existing file is never overwritten, and the path is checked before streaming).
• Importing the LJM wrapper prints to stdout if the native library is missing. The server redirects that import so the MCP stdio channel is never corrupted, and reports a clear install hint instead.

Hardware verification