Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@rigol capture a screenshot of the scope"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Control a Rigol DS1054Z oscilloscope on USB, from Claude Code sessions (MCP server) and from people
and scripts (the rigol CLI). Linux only: it drives the kernel usbtmc driver directly, working
around the scope's USB firmware bugs.
uv sync
# once, as root: let the plugdev group open the scope
echo 'KERNEL=="usbtmc[0-9]*", ATTRS{idVendor}=="1ab1", GROUP="plugdev", MODE="0660"' | sudo tee /etc/udev/rules.d/99-rigol.rules
sudo udevadm control --reload && sudo udevadm trigger --subsystem-match=usbmisc
# make the tools available to every Claude Code session on this machine
claude mcp add --scope user rigol -- "$PWD/.venv/bin/rigol-mcp"Related MCP server: Rigol DHO824 MCP Server
Numbers take SI suffixes (500m, 20ns, 2.5V). Changes print each setting's requested, before
and after values. The scope snaps to the steps it supports, so "after" is what you got.
# reading
uv run rigol idn
uv run rigol screenshot # prints the PNG's path
uv run rigol settings # channels, timebase, trigger, acquisition (JSON)
uv run rigol measure 1 VPP FREQ VRMS # null = the scope couldn't measure it
uv run rigol waveform 1 # on-screen trace → CSV, plus summary stats
uv run rigol query ':TRIGger:STATus?' # any single read-only SCPI query
uv run rigol save-setup # whole setup → file
# changing (only the options given)
uv run rigol channel 2 --on --scale 500m --coupling AC --probe 10x
uv run rigol timebase --scale 20ns --offset 0
uv run rigol trigger --source 1 --slope rising --level 1.5 --sweep normal
uv run rigol acquisition --type averages --averages 16 --memory-depth 12k
uv run rigol run | stop | force
uv run rigol single --timeout 5 # arm, and wait up to 5 s for the capture
# whole-setup changes: each saves a backup first and prints its path
uv run rigol restore-setup FILE
uv run rigol autoscale --yes
uv run rigol reset --yes
uv run rigol write ':CHANnel1:VERNier ON'
uv run rigol clear-measurements [--item N]Captures and backups go to ~/.local/share/rigol-remote/captures/ (override with
RIGOL_CAPTURE_DIR). The scope is found automatically; set RIGOL_ADDR=/dev/usbtmcN to pick one
explicitly. Concurrent users (sessions, scripts) take turns via a lock on the device.
Tier | Tools |
Read-only |
|
Changes (undoable: pass the returned before-values back) |
|
Destructive |
|
The tools carry MCP read-only/destructive hints. In Claude Code you can allow the read-only ones
without prompting, e.g. "mcp__rigol__screenshot" in permissions.allow, and keep the rest behind
approval.
uv run pytest # unit tests, no scope needed
uv run pytest -m hardware # read-only tests against the real scope
uv run pytest -m hardware_write # changes settings on the scope, then restores the saved setupLet the scope choose scales, timebase and trigger for the connected signals, like the AUTO key. It replaces the user's setup, so it is backed up first. Returns the backup path and the new settings.
Inputs: no named parameters.
Remove one item (1-5, left to right) or, by default, all items from the measurement bar at the bottom of the scope's screen. That includes items the user added.
Inputs: item.
Change how the scope acquires; only the parameters given are changed. averages (a power of two, 2-1024) applies in AVERAGES mode. Valid memory depths depend on how many channels are on (1 channel: 12000 to 24000000; 2: 6000 to 12000000; 3-4: 3000 to 6000000). Returns requested/before/after.
Inputs: averages, memory_depth, acquisition_type.
Change one channel's settings; only the parameters given are changed. channel: 1-4 or CHAN1-CHAN4. probe_ratio must match the probe's attenuation switch (1, 10, ...); it rescales the V/div. Returns requested/before/after for each setting.
Inputs: channel (required), invert, display, bw_limit, +4 more.
Change the horizontal settings; only the parameters given are changed. offset_s moves the trigger point. Returns requested/before/after for each setting.
Inputs: mode, offset_s, scale_s_per_div.
Change the edge-trigger settings; only the parameters given are changed. source: 1-4, CHAN1-CHAN4 or AC (mains). Giving source, slope or level switches the trigger to EDGE mode. sweep AUTO free-runs without a trigger, NORMAL waits for one. Returns requested/before/after.
Inputs: slope, sweep, source, level_v, +2 more.
Channel (on/off, V/div, offset, coupling, probe ratio), timebase, trigger and acquisition settings. Values are SI: volts, seconds, samples per second.
Inputs: no named parameters.
Read one channel's on-screen trace (1200 points), save it as CSV (time_s,volts) and return
summary statistics plus the CSV path. clipped: true means the trace ran off screen, so its
extremes are unreliable. channel: 1-4 or CHAN1-CHAN4.
Inputs: channel (required).
Manufacturer, model, serial number and firmware of the attached scope.
Inputs: no named parameters.
The scope's automatic measurements on one channel. channel: 1-4 or CHAN1-CHAN4. items: any of FREQ, FTIM, MAR, MPAR, NDUT, NEDG, NPUL, NSLEW, NWID, OVER, PDUT, PEDG, PER, PPUL, PRES, PSLEW, PVRMS, PWID, RTIM, TVMAX, TVMIN, VAMP, VARI, VAVG, VBAS, VLOW, VMAX, VMID, VMIN, VPP, VRMS, VTOP, VUP (long forms like FREQuency also accepted). A value of null means the scope could not measure it. Side effect: each item is added to the measurement bar on the scope's screen (it shows at most 5, so older ones scroll off).
Inputs: items (required), channel (required).
Reset the scope to factory defaults (*RST). The setup is backed up first. Returns the backup path and the new settings.
Inputs: no named parameters.
Restore a whole setup saved by save_setup (or a backup from another tool). The current setup is backed up first. Takes about 6 s. Returns the backup path and the resulting settings.
Inputs: path (required).
Start (run) or freeze (stop) acquisition, arm one acquisition (single), or force a trigger. run and single overwrite a stopped capture. For single, wait_s waits up to that long for it to complete. Returns the trigger status (RUN, WAIT, TD, AUTO or STOP) and sweep mode.
Inputs: action (required), wait_s.
Save the scope's whole setup to a file and return its path; restore_setup(path) brings it back.
Inputs: no named parameters.
Send one read-only SCPI query (its header must end in '?', e.g. ':TRIGger:STATus?' or ':CHANnel1:SCALe?') and return the reply. Commands that change settings are refused.
Inputs: command (required). Returns: result
Send one raw SCPI command that expects no reply, for anything the other tools don't cover (e.g. ':CHANnel1:VERNier ON'). Use scpi_query for queries. The setup is backed up first. Fails if the scope reports an error.
Inputs: command (required).
Capture the scope's screen (800x480 PNG). Also saved to disk; the path is returned.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 17 tools
Most tools have clearly distinct purposes (screenshot, identify, measure, get_waveform, the configure_* family). The only notable overlap is the generic scpi_query/scpi_write pair against the dedicated configure/get tools, but the descriptions explicitly frame them as escape hatches for uncovered functionality, which mitigates confusion.
The set follows a mostly verb/verb_noun imperative pattern (get_settings, save_setup, restore_setup, configure_channel, clear_measurements, run_control) with readable single-word verbs (identify, measure, autoscale, reset). Minor deviations like the noun 'screenshot' and the prefix-ordered scpi_query/scpi_write keep it from being perfectly uniform.
17 tools is slightly above the ideal 3-15 range but each one maps to a genuine, non-redundant oscilloscope capability (acquisition, trigger, channels, measurements, waveform, setup lifecycle). Nothing feels padded for a full instrument-control surface.
The surface covers the core lifecycle well: identity, settings/waveform/measurement reads, channel/timebase/trigger/acquisition configuration, run control, setup backup/restore, autoscale and reset. Gaps like math/FFT, cursors, or non-edge trigger modes are real but workable through the scpi_write escape hatch and setup backup protection.
Use your Mac, Windows or Linux computer from ChatGPT, Claude or Codex: files, commands, documents.
Work on your own computers from Claude or ChatGPT: run commands, edit and search files.
Work with your DealSharp negotiation projects, documents and tools from Claude.
Amazon Seller Central, Ads and Vendor Central in ChatGPT & Claude. 106 tools; writes need approval.
