ltspice-mcp

Work in progress. Core functionality is usable but expect rough edges and breaking changes.

An MCP server that connects LLM assistants (Claude, and any other MCP client) to real circuit simulation: LTspice and ngspice, plus direct editing of LTspice .asc schematics. Simulation results come back as structured numbers — cutoff frequencies, overshoot, phase margin, rise times, and per-device small-signal operating-point parameters (gm, gds, vth, …) read back by name — so the assistant can design, verify, and iterate on circuits in the same files you open in LTspice, without ever hand-parsing a rawfile. Built on spicelib.

Quick start

In Claude Code, install the plugin:

/plugin marketplace add cognitohazard/ltspice-mcp
/plugin install ltspice-mcp

You also need LTspice or ngspice on the host (auto-detected on Windows, Linux, and macOS; on WSL set the LTspice path explicitly — WSL notes). Netlist editing (.cir/.net) works with no simulator at all; .asc schematic editing needs LTspice's .asy symbol libraries. uv is required; the server itself is fetched from PyPI on first use.

Manual install (any MCP client)

Install the server, then point your client at it:

uv tool install ltspice-mcp     # or: pip install ltspice-mcp / pipx install ltspice-mcp

Claude Code — one command (drop -s project to install it globally):

claude mcp add -s project ltspice -- ltspice-mcp

Other clients — Claude Desktop, Cursor, Windsurf, Gemini CLI, Continue, Cline, Zed and others — add this mcpServers stanza to the client's MCP config file (each client documents its own path):

{
  "mcpServers": {
    "ltspice": { "command": "ltspice-mcp", "args": [] }
  }
}

Python 3.11+ required. Verify with ltspice-mcp --help. The same server is also published under two alias names — circuit-mcp and ngspice-mcp — so uvx circuit-mcp / uvx ngspice-mcp are drop-in equivalents of uvx ltspice-mcp if one of those names is more discoverable for you.

Web clients (claude.ai, ChatGPT) need a stdio→HTTP bridge such as mcp-proxy — only expose this server on a network you fully control, since it writes files and spawns processes inside allowed_paths.

A Claude Desktop extension is also available: build the .mcpb in packaging/mcpb/ and drag it onto Claude Desktop for a one-click install with a native folder picker for your circuits directory. Like the plugin, it wraps the PyPI package and needs uv and a simulator on the host (it does not bundle LTspice or ngspice).

Related MCP server: SPICEBridge

Using it

Once connected, you ask for circuit work in plain language. The assistant designs the circuit and decides what to measure; the server runs the simulator, parses the binary output, and hands back the numbers. It reports what the run produced, the simulator's own warnings included, and leaves the call on whether a result is good to you and the assistant.

"Bias this NMOS common-source stage into saturation at the target drain current and report gm/ID."

The assistant writes the netlist, solves the bias point on LTspice, and reads the device's operating point back by name — drain current, gm, gds, VDS against VDSAT to confirm it's in saturation, and the gm/ID that analog designers size to. If the bias is off, it nudges the gate reference or W/L and re-runs, a couple of seconds per pass.

Other requests that work the same way:

• "What's the overshoot and settling time of this regulator's step response?" — runs a transient analysis and measures both from the waveform, plus rise time, ringing frequency, and the final value.

• "Run a 200-run Monte Carlo with 5% resistors and tell me the output spread." — perturbs components per run, simulates the batch, and reports mean, sigma, and worst-case values per measurement.

• "Sweep the load from 100 Ω to 10 kΩ and find where efficiency drops." — parameter sweep with per-run results.

• "Characterize this NMOS: gm and gm/ID vs VGS." — writes a .dc Vgs deck with .save @m1[gm] @m1[id], runs it on ngspice, and returns the gm/ID table as one CSV (no .control block, no rawfile parsing).

• "Find an N-channel power MOSFET for a low-side switch and measure the on-state drop." — searches the loaded libraries for a part (find_model), drops it into a pulsed-gate transient, and reads Vds(on) and load current back from the .meas results.

• "Build this differential pair as a schematic I can open in LTspice." — places and wires the components into a real .asc, with orthogonal routing and pin-collision checks.

• "Is this loop stable?" — AC analysis of the loop gain; reports phase and gain margin at every crossover, not just the first.

• "What's the resonant frequency and Q of this series RLC?" — runs an AC sweep and reports each peak's center frequency, Q, and −3 dB bandwidth.

The warning rides with the number it affects. A simulator like ngspice can print "singular matrix" once, deep in a log you'd never open, then finish the run and write perfectly plausible numbers anyway — read them by hand and nothing looks off. Ask the server for one of those numbers and the buried line comes attached to it, in an observations field right next to the value, so the failure surfaces where you're already looking instead of where it's easy to scroll past.

Co-design on the same files

Everything operates on ordinary LTspice and SPICE files, so the work passes back and forth between you and the assistant instead of living inside a chat:

• Sketch a schematic in LTspice, then hand it over: "what's the bias point?", "why doesn't the output move?", "add compensation and check the phase margin."

• Or the reverse: the assistant designs and verifies the circuit and writes the .asc; you open it in LTspice, inspect it, and tweak by hand. Your manual edits are simply the file's new state — the assistant picks up from there on the next request.

• Changes can flow either direction mid-design: adjust a value in the GUI and ask for re-verification, or have the assistant sweep a change you're considering before you commit to it.

What it does

Simulation and measurement. Runs LTspice or ngspice and parses the binary output directly. Measurements are computed server-side and returned as numbers: time-domain (rise/fall, overshoot, settling, delay, period/duty/jitter, RMS, THD), frequency-domain (filter cutoffs and roll-off, gain and phase at any frequency, stability margins, resonance peaks with Q, integrated noise), DC operating points, and .MEAS directive results including the ones that failed. Per-device small-signal operating-point parameters (gm, gds, vth, …) come back by name on both simulators — LTspice via an auto-added .options logopinfo block in the log, ngspice via .save @dev[param] traces. Read the set across a .dc sweep as a gm/ID table with export_waveform, or a single bias point with operating_point (address them as m1.gm / @m1[gm], no rawfile parsing).

Schematic and netlist editing. Creates and edits real LTspice .asc files — place components, wire pins, label nets — with validation before anything is written: wiring that would collide with a pin, overlap a junction, or run diagonally is refused, and every edit returns warnings about floating pins or dangling labels. A session's edits can be reverted. Plain netlists (.cir/.net) get the same operations at text level, plus a static validation pass that catches malformed cards before a simulation is spent.

Sweeps and Monte Carlo. Multi-dimensional parameter sweeps and Monte Carlo with per-component tolerances, .MODEL process variation, and Pelgrom W·L device mismatch. Per-measurement statistics are aggregated across runs, and any single run can be pulled out and analyzed like a standalone simulation.

Jobs and trust. Simulations run as cancellable jobs with timeouts and a concurrency cap; long runs return a job ID immediately and job state survives a server restart. Results report facts, not verdicts: a completed run carries the simulator's own warnings, measurements that produced nothing, and extreme node values as structured observations. Judging whether a result is trustworthy is left to the model reading it.

Supported simulators

Configuration

Works with defaults out of the box. To customize, copy ltspice-mcp.example.toml to ltspice-mcp.toml; any setting can be overridden with an LTSPICE_MCP_-prefixed environment variable, and --config PATH or LTSPICE_MCP_CONFIG picks the file. Key options:

[simulator]
default = "ltspice"      # ltspice, ngspice, qspice, xyce (null = auto-detect)
path = ""                # explicit executable path (required on WSL)
ngbehavior = "hsa"       # ngspice compat mode; unset = spicelib default, "hsa" fixes sectioned .lib corner select

[security]
allowed_paths = ["."]    # sandbox: only these directories are accessible

[simulation]
# max_parallel = 4       # default: number of CPU cores, capped at 8
timeout = 300.0          # seconds

[tools]
profile = "full"         # or "agentic"

[state]
persist_jobs = true

See src/ltspice_mcp/config.py for the full option list ([analysis], [schematic], [logging], ...).

On WSL, LTspice.exe runs via Windows interop (not Wine), and spicelib can't auto-detect it across the WSL boundary. Set the Windows-side path explicitly:

[simulator]
path = "/mnt/c/Program Files/ADI/LTspice/LTspice.exe"

Simulation output is automatically redirected to a Windows temp directory: LTspice's .MEAS results go through SQLite .db files that fail on UNC paths (\\wsl.localhost\...), and without the redirect measurement data silently disappears from the logs.

.asy symbol paths for .asc editing are auto-detected on Windows and WSL; override with [schematic] symbol_paths or LTSPICE_MCP_SYMBOL_PATHS.

Tool profiles

The agentic profile drops netlist-editing wrappers and library session management — work a capable agent does through direct file edits — and keeps simulation lifecycle, binary .raw parsing, batch orchestration, and the .asc geometry tools. The skills/ directory (skills/ltspice/SKILL.md, skills/ngspice/SKILL.md) contains the domain knowledge that pairs with it: copy the relevant skill into your client's persistent-instructions location.

Where it runs. The server shells out to a local LTspice/ngspice and reads circuit files from disk, so it must run where the simulator and the files are. Two setups work: a local MCP host (Claude Desktop, Claude Code, Cursor, Gemini CLI, Codex, …) on your own machine, or a browser-based cloud agent whose sandbox can install ngspice and register the server (verified with Claude). LTspice is local-only (a Windows app); ngspice is open-source and works in either place. Consumer web chat with no sandbox has no simulator and no file access, so it can't run this server directly; bridge it to a machine you control (e.g. mcp-proxy) if you want that UI.

Under the hood: the tool-level loop

What the assistant actually does for "design a 1 kHz RC low-pass and verify it". It writes the netlist (R=1k, C=159.155n → fc = 1 kHz):

* rc.cir — RC low-pass
V1 in 0 AC 1
R1 in out 1k
C1 out 0 159.155n
.ac dec 50 1 1Meg
.end

then drives three tools:

validate_netlist(path="rc.cir")
  → OK: directives valid, element arities check out — safe to simulate

run_simulation(netlist="rc.cir")
  → {"job_id": "sim_a3f1", "status": "completed", "raw_file": ".../rc.raw", ...}

bode_metrics(raw_file=".../rc.raw", signal="V(out)", mode="filter")

and gets back scalars, not a plot:

{
  "signal": "V(out)",
  "filter_type": "lowpass",
  "passband_gain_db": 0.0,
  "passband_ripple_db": 0.02,
  "cutoff_low_hz": null,
  "cutoff_high_hz": 1000.4,
  "stopband_rejection_db": 59.97,
  "rolloff_slope_db_per_decade": -19.9,
  "estimated_order": 1,
  "warnings": []
}

(abridged — the full response also includes passband bounds and transition bandwidth)

Off-target → set_component_value, re-run, re-measure. Long simulations return a job ID instead of blocking; check_job/cancel_job manage them. Job metadata persists in per-circuit sidecars ({dir}/.ltspice-mcp/jobs/ — add .ltspice-mcp/ to your .gitignore), and MCP resources (spice://results/..., spice://netlists/..., spice://config) expose jobs, signals, measurements, and config for browsing.

Every tool declares MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint); data-returning tools declare an outputSchema for structuredContent introspection.

Development

uv sync                        # install runtime + dev dependencies
uv run pytest tests/ -v        # tests
uv run pyright                 # type checking
uv run ruff check src/ tests/  # lint
uv run ltspice-mcp             # run the server (stdio)

More: docs/DESIGN.md (scope, architecture, non-goals) and docs/spice_lex.md (SPICE parser internals).

License

GPL-3.0