SRS Lock-in Amplifier — MCP Server

Let an AI agent read X/Y/R/θ, set the reference frequency, harmonic and phase, drive level, sensitivity and time constant, run auto phase/gain, and record bounded frequency sweeps on Stanford Research Systems DSP lock-in amplifiers, both the classic SR810/SR830 and the current SR860/SR865A.

Try it without hardware

uvx labmcp-srs-lockin --simulate --check                       # SR830 simulator
uvx labmcp-srs-lockin --simulate --option model=sr860 --check  # SR860 simulator

The simulated sample is a resonator (10 kHz, Q = 50) driven by the sine output, so a frequency_sweep from 9 to 11 kHz shows a resonance peak.

Connect your lock-in

The model is detected from *IDN? (override with --option model=sr830|sr810|sr860|sr865a).

1. GPIB (all models): set the GPIB address in the setup menu (SR830 factory default 8). Install labmcp[visa] (included) and a VISA backend (pyvisa-py works with most USB-GPIB adapters; NI-VISA with ?backend=@ivi).
2. RS-232: set baud rate and parity on the instrument (SR830 default 9600 baud, no parity; the word length is always 8 bits). The SR830 only answers on the interface selected by OUTX; the server sends OUTX 0 automatically on serial connections and OUTX 1 on GPIB. The SR830 ends RS-232 replies with CR. For an SR86x on RS-232, pass --option model=sr860 (replies end with LF) or set the SR86x transmit terminator to CR.
3. SR86x USB / Ethernet: use the USBTMC or VXI-11 VISA resource (TCPIP0::<ip>::inst0::INSTR; enable VXI-11 and add your PC to the trusted IP range in the system menu).
4. Test the connection:

   uvx labmcp-srs-lockin --address GPIB0::8::INSTR --check
   uvx labmcp-srs-lockin --address /dev/ttyUSB0 --check                      # SR830 RS-232
   uvx labmcp-srs-lockin --address "serial://COM3?baudrate=19200" --check    # non-default baud
   uvx labmcp-srs-lockin --address TCPIP0::192.168.1.20::inst0::INSTR --check  # SR86x LAN
   

   Use --option interface=gpib if you reach an SR830's GPIB port through a raw TCP adapter.

Add to your MCP client

Claude Code

claude mcp add lockin -- uvx labmcp-srs-lockin --address GPIB0::8::INSTR

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

{
  "mcpServers": {
    "lockin": {
      "command": "uvx",
      "args": ["labmcp-srs-lockin", "--address", "GPIB0::8::INSTR"]
    }
  }
}

Add --read-only to allow reading outputs and settings but block every setting change and the sine-output drive. For other clients, generate the snippet with uvx labmcp config srs-lockin --address GPIB0::8::INSTR --client vscode (also cursor, codex, claude-code).

Tools

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

Safety limits

Override at launch: --limit max_amplitude_v=0.1 --limit max_sweep_duration_s=1800.

Example prompts

• "Set the lock-in to 13.7 Hz internal reference, 100 mV drive, 1 s time constant at 24 dB/oct, and tell me X and Y once it has settled."
• "Auto-phase, then read R and θ ten times and give me the mean and standard deviation of R."
• "Sweep the drive from 30 to 34 kHz in 201 log-spaced steps at 10 mV and find the resonance frequency and Q of my tuning fork."
• "Switch to the current input with 100 MΩ gain, DC coupling, and pick the sensitivity automatically."
• "We're done - turn the sine output down to the minimum."

Notes

• The sine output is the hazard. It drives your sample (heating, electromigration, piezo stress). set_amplitude checks max_amplitude_v; set_amplitude_minimum is the closest thing to "output off": the SR810/SR830 cannot go below 4 mV rms, so disconnect the cable if the sample must see no drive. On the SR86x it sets 1 nV rms and a 0 V DC level (SOFF). The SR830 powers up and resets (*RST) to 1.000 V rms.
• Every set command is followed by *ESR?; if the lock-in flags an execution error (parameter out of range, e.g. FREQ in external-reference mode) or an illegal command, the tool reports it instead of silently continuing.
• set_sensitivity picks the smallest range that is at least the requested full scale (never a smaller one, which would overload). set_time_constant picks the nearest available value; the SR830 may raise it to the minimum allowed for the slope/reserve, and the returned settings show the value actually used.
• read_outputs returns unit = V for voltage inputs and A for the current input (1 V ↔ 1 µA sensitivity mapping from the manuals). The SR830 manual (SR530-compatibility QX/QY commands, ch. 5) states that outputs are returned in amps when the current input is selected; the SR86x manual does not say explicitly — please report what your instrument returns.
• overloads comes from LIAS?, whose bits are latched since the last read; the first reading after a transient may show an overload that has already cleared.
• frequency_sweep waits the manual's 99 % settling time at each point (5/7/9/10 time constants for 6/12/18/24 dB/oct) unless settle_time_constants is given, and restores the starting frequency afterwards. Calling set_amplitude_minimum stops a running sweep at the next point (also one that is still reading its settings). One sweep can take at most 3640 s (the tool's timeout minus a margin) whatever max_sweep_duration_s is: longer estimates are refused, and a sweep running slower than estimated stops early and returns the points measured so far (completed: false). save_path must end in .csv and must not exist yet (an existing file is never overwritten); it is checked before the sweep starts.
• set_amplitude_minimum tries both the amplitude and (SR86x) the DC level even if one of them fails, then reports every failure.
• auto_gain maps to AGAN (SR830) or ASCL (SR86x); auto_range maps to ARSV (SR830 auto reserve) or ARNG (SR86x auto input range). SR830 AGAN does nothing when the time constant is above 1 s.
• Data-buffer, streaming and FFT functions are not exposed. The SR850, SR844 (RF) and SR865 (non-A) are not included because they were not verified; an explicit --option model=... lets you try a compatible model at your own risk.

Hardware verification