ClaudeCodeMod

All shelves / MCP servers

Ocean Insight Spectrometer

io.github.K-Dense-AI/labmcp-ocean-spectrometer · 19 stars · Python · Apache-2.0

MCP server MCP server for Ocean Insight / Ocean Optics spectrometers (python-seabreeze).

Install

The repo has no one-line install. Follow its README.

Published as

  • PyPIlabmcp-ocean-spectrometerstdio

From the server's entry in the official MCP Registry.

Open the repo

Configuration

NameSet asWhat it is
LABMCP_ADDRESSOptionalEnv varInstrument address, e.g. serial:///dev/ttyUSB0 or tcp://192.168.1.50:5025
LABMCP_SIMULATEOptionalEnv varSet to 1 to use the built-in simulator (no hardware)
LABMCP_READ_ONLYOptionalEnv varSet to 1 to disable all state-changing tools
LABMCP_LIMITSOptionalEnv varSafety limit overrides, e.g. max_temperature_c=80

Files

servers/chemistry/ocean-spectrometer/README.md

Ocean Insight Spectrometer — MCP Server

Let an AI agent acquire spectra, set or auto-tune the integration time, store dark and reference spectra, and measure absorbance and transmittance with peak positions and FWHM on Ocean Insight (Ocean Optics) spectrometers. It works through the open-source python-seabreeze library.

Package labmcp-ocean-spectrometer
Instruments Every model python-seabreeze supports: USB2000+, USB2000, USB4000, USB650, Flame-S, Flame-NIR/X, HR2000/HR2000+, HR4000, Maya2000/Pro/LSL, QE65000, QE Pro, STS, NIRQuest256/512, Jaz, Torus, Spark, Apex, Ventana, HDX, ADC1000-USB, HR2/4/6, SR2/4/6, ST-VIS (support table; some models need a specific backend)
Interfaces USB
Protocol python-seabreeze 2.x (docs, source), which implements Ocean's SeaBreeze (OOI / OBP USB protocols). TEC details: Ocean QE Pro Installation and Operation Manual, MNL-0000
Status 🧪 simulated: tested against a physically modelled fake spectrometer with the seabreeze Spectrometer interface, not yet verified on hardware. Report a hardware test

Try it without hardware

uvx labmcp-ocean-spectrometer --simulate --check
uvx labmcp-ocean-spectrometer --simulate --option sim_source=hg-ar --check      # Hg-Ar calibration lamp
uvx labmcp-ocean-spectrometer --simulate --option sim_model=QE-PRO --check      # TE-cooled, 1044 pixels

The simulator models a USB2000+ (or QE Pro) looking at a tungsten-halogen lamp, a white LED or a Hg-Ar lamp. It includes shot and read noise, dark offset and dark current, electric-dark pixels, detector nonlinearity, saturation, and the stale first spectrum after an integration-time change. In simulation a "virtual user" blocks the beam before store_dark_reference, inserts a blank before store_reference, and inserts a dye sample (absorbance band at 520 nm, 0.8 AU) before measure_absorbance/measure_transmittance. With real hardware your scientist does these steps.

Connect your spectrometer

  1. Drivers / permissions (once per computer), as documented by python-seabreeze:
    uvx --from seabreeze seabreeze_os_setup      # or, inside the environment: seabreeze_os_setup
    • Linux: this installs the udev rules (/etc/udev/rules.d/10-oceanoptics.rules, needs sudo). Re-plug the spectrometer afterwards.
    • Windows: this installs the USB driver files. The pyseabreeze backend also needs libusb-1.0.dll (see the seabreeze README).
    • macOS: works out of the box.
  2. Close OceanView (or any other program using the spectrometer). Only one program can open it.
  3. Test the connection:
    uvx labmcp-ocean-spectrometer --check                        # the only connected spectrometer
    uvx labmcp-ocean-spectrometer --address USB2+H01234 --check  # choose by serial number
    uvx labmcp-ocean-spectrometer --option backend=pyseabreeze --check
    --address is the spectrometer's serial number and is only needed when several are connected (list_spectrometers shows them). The default backend is cseabreeze (compiled libseabreeze); if it can't load, seabreeze falls back to the pure-Python pyseabreeze (pyusb).

On connect the server sets the integration time to 10 ms (--option integration_ms=…), clamped to the device range. It has to set it, because seabreeze cannot read the integration time back.

Add to your MCP client

Claude Code

claude mcp add spectrometer -- uvx labmcp-ocean-spectrometer

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

{
  "mcpServers": {
    "spectrometer": {
      "command": "uvx",
      "args": ["labmcp-ocean-spectrometer", "--address", "USB2+H01234"]
    }
  }
}

Add --read-only to allow acquisitions and absorbance measurements with the present settings while blocking integration-time changes, reference storage and TEC control. For other clients, generate the snippet with uvx labmcp config ocean-spectrometer --client vscode (also cursor, codex, claude-code).

Tools

Tool Kind Description
acquire_spectrum 👁 read Acquire an intensity spectrum (raw counts) with optional averaging, boxcar smoothing and corrections. Returns downsampled (wavelength, counts), summary statistics, the most prominent peaks with FWHM, and a saturation check; save_path writes the full spectrum to a new .csv file (an existing file is never overwritten).
auto_integration_time 🎛 control Adjust the integration time until the brightest raw pixel (optionally within a wavelength window) is within the target band of saturation (default 70-85 %). Stays within the device limits and the max_integration_time_ms safety limit. Re-take dark/reference afterwards.
detector_cooling_off 🛑 safety Switch the detector thermo-electric cooler off (the detector warms to ambient). Does nothing on spectrometers without a TEC.
find_peaks 👁 read Find peaks (or dips) with position, height, prominence and FWHM in the most recent spectrum (intensity, absorbance or transmittance). Acquires a fresh intensity spectrum if none has been taken yet.
get_command_log 👁 read Return the most recent raw commands sent to / replies received from the instrument (newest last). Useful for debugging and for recording what was done.
get_connection_info 👁 read Report which instrument is connected (identity, address, simulated or real), whether the server is read-only, and the active safety limits. Call this first.
get_device_info 👁 read Model, serial, pixel count, wavelength range, integration-time limits, saturation level, supported corrections, TEC presence, and which dark/reference spectra are stored.
list_spectrometers 👁 read List the Ocean spectrometers seabreeze can see (model, serial number, whether open) and which one this server is connected to. Use a serial number as --address to pick one.
measure_absorbance 👁 read Measure the absorbance spectrum A = -log10((S - D) / (R - D)) of the sample now in the beam, using the stored dark D and reference R (same settings). Ask the user to insert the sample first. Also returns values at wavelengths_nm and the absorbance maxima.
measure_transmittance 👁 read Measure the transmittance spectrum %T = 100 (S - D) / (R - D) of the sample now in the beam, using the stored dark and reference. Ask the user to insert the sample first. Also returns values at wavelengths_nm and the deepest transmission dips.
read_detector_temperature 👁 read Read the detector temperature from the thermo-electric cooler (TE-cooled models such as the QE Pro, when seabreeze exposes the thermo_electric feature for them).
reconnect 🛑 safety Close and re-open the connection to the instrument (e.g. after it was power cycled or a cable was re-plugged).
set_detector_cooling ⚠️ hazard Enable the detector thermo-electric cooler at setpoint_c (TE-cooled models only). The detector takes minutes to settle; dark current (and noise) drops as it cools. The TEC can only cool to roughly 15-40 °C below ambient (QE Pro manual). Checked against min_tec_setpoint_c.
set_integration_time 🎛 control Set the detector integration time (ms), within the device limits from get_device_info. Stored dark/reference spectra become unusable until re-taken at the new time.
store_dark_reference 🎛 control Record and store a DARK spectrum (kept in memory) for absorbance/transmittance. BEFORE calling, ask the user to BLOCK THE LIGHT PATH (close the shutter, switch the lamp off or cap the fiber) and wait for confirmation. Uses the present integration time.
store_reference 🎛 control Record and store the REFERENCE (100 % transmission) spectrum in memory. BEFORE calling, ask the user to put the BLANK (solvent-filled or empty cuvette) in the beam with the light on, and wait for confirmation. Use the same settings as the dark.

get_connection_info, get_command_log and reconnect are built into every LabMCP server.

Safety limits

Limit Default Meaning
max_integration_time_ms 10000 ms Longest integration time an agent may set (also caps auto_integration_time)
max_acquisition_duration_s 300 s Longest single acquisition (scans × integration time)
min_tec_setpoint_c −20 °C Coldest TEC setpoint (TE-cooled models). The QE Pro's temperature readback is linearised for −20…40 °C

Override at launch: --limit max_integration_time_ms=60000. Whatever the limits say, one acquisition (scans × integration time, plus the spectrum discarded after an integration-time change) must fit in 540 s, the time one tool call may take; auto_integration_time stops early for the same reason and says so.

Example prompts

  • "Auto-adjust the integration time on the lamp, then take a 20-scan averaged spectrum and tell me where the peak is and how wide it is."
  • "I'm going to measure a Beer–Lambert series. Walk me through dark, blank and each of my five standards, and report the absorbance at 520 nm for each."
  • "Here's the Hg-Ar lamp. Find the lines between 400 and 600 nm and tell me how far they are from 404.66, 435.83 and 546.07 nm."
  • "Measure the transmittance of this filter from 400 to 700 nm and save it to ~/data/filter_T.csv."
  • "Cool the QE Pro to −10 °C and tell me when the detector temperature has settled."

Notes

  • Raw counts. Intensities are detector counts, not irradiance. Radiometric calibration is out of scope.
  • Saturation is judged on the raw ADC counts of every averaged scan: a pixel is saturated at ≥ 98 % of max_intensity. Saturated pixels are flagged, and they're excluded from absorbance/transmittance (with boxcar smoothing, so are the pixels within the boxcar half width of them). seabreeze notes that some detectors saturate below max_intensity, so keep peaks at 70–85 %.
  • Corrections. correct_dark_counts subtracts the mean of the optically masked electric-dark pixels; correct_nonlinearity applies the EEPROM polynomial. The driver reads raw spectra and applies both corrections with the same formula as seabreeze.spectrometers.Spectrometer.intensities, so saturation can still be detected on raw data. A stored dark (store_dark_reference) is still required for absorbance, because electric-dark correction doesn't remove fixed-pattern dark signal.
  • Absorbance is A = −log10((S − D)/(R − D)). Pixels where the reference has less than 0.5 % of its peak signal above dark, or where the reference or sample is saturated, are returned as null. The dark, reference and sample must share integration time, boxcar and corrections; the server enforces this and asks you to re-take references when they differ.
  • First spectrum after an integration-time change is discarded. It may have been integrated partly with the old setting; the QE Pro manual notes that it returns the most recently completed spectrum.
  • Boxcar follows OceanView's convention: boxcar_half_width = pixels averaged on each side. Peaks are found on the full-resolution spectrum. The returned curve is bin-averaged down to max_points.
  • TEC tools use seabreeze's thermo_electric feature (enable_tec, set_temperature_setpoint_degrees_celsius, read_temperature_degrees_celsius) and only work when the backend exposes it for your model. In pyseabreeze 2.11 that is only the NIRQuest512. For the QE Pro it depends on libseabreeze (cseabreeze); this is unverified on hardware. The QE Pro manual says the TEC cools only ~15–40 °C below ambient.
  • Not exposed: shutter and lamp (light_source/strobe_lamp) control, trigger modes, and irradiance calibration. seabreeze documents these features, but device support isn't specified and several aren't implemented in pyseabreeze, so they could not be verified.
  • Peaks next to invalid pixels. Invalid (null) pixels are treated as unknown: a pixel next to one is not reported as a peak, and prominence and FWHM do not extend across them. When invalid pixels lie inside the requested range (for example the centre of a band too strong to measure), absorbance/transmittance results carry a warning.
  • Dark and reference spectra live in the server's memory and are lost on reconnect or restart. Save data with save_path: it must end in .csv, missing folders are created, and an existing file is never overwritten (choose a new name). The path is checked before the spectrum is acquired.

Hardware verification

Model Firmware Interface Verified by Date
none yet: be the first

Facts

Kind
MCP server
Repo
K-Dense-AI/lab-instrument-mcps
Group
Uncategorized
Stars
19
License
Apache-2.0
Language
Python
Last push
2026-09-26
Forks
1
MCP Registry
io.github.K-Dense-AI/labmcp-ocean-spectrometer
Homepage
www.k-dense.ai
Topics
ai-agents, instrument-control, lab-automation, laboratory, mcp, mcp-server, model-context-protocol, python, scientific-instruments, self-driving-labs

More on this shelf

  1. 1REAio.github.morluto/reaReverse engineer anything from your terminal or agent with one CLI and MCP server.125.3k
  2. 2com.browser-use/browser-usecom.browser-use/browser-useControl a real Chrome browser to complete any task: fill forms, extract data, book flights.117.7k
  3. 3World Monitorapp.worldmonitor/mcpLive markets, conflicts, country risk, chokepoints, energy, and China decision signals. 93 tools.88.3k
  4. 4Scrapling MCP Serverd4vinci/scrapling🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scraplingdev86.9k
  5. 5Everythingmodelcontextprotocol/serversThis MCP server attempts to exercise all the features of the MCP protocol. It is not intended to be a useful server, but rather a test server for builders of MCP clients. It implements prompts, tools, resources, sampling, and more to showcase MCP capabilities.85.8k
  6. 6Fetchmodelcontextprotocol/serversA Model Context Protocol server that provides web content fetching capabilities. This server enables LLMs to retrieve and process content from web pages, converting HTML to markdown for easier consumption.85.8k