Skip to main content
Glama
spicelib-mcp

spicelib-mcp

by lucasgerads

spicelib-mcp

A thin MCP server that wraps spicelib for circuit simulation. The heavy lifting, such as simulator invocation, .raw file parsing, and multi-simulator abstraction, is done by spicelib, which deserves most of the credit. This project is purely a transport layer that exposes spicelib's functionality as MCP tools for use with Claude Code (and other coding agents).

Primary use case: behavioral model fitting — run simulations, compare against real measurements, iterate on SPICE models.

Tools

  • run_ac_analysis — AC frequency sweep, returns magnitude/phase data

  • run_transient — transient simulation, returns time-domain waveforms

  • run_dc_op — DC operating point analysis

  • run_sweep — run multiple simulations in parallel, sweeping component values

Results are saved as .npz files alongside the netlist for further analysis and plotting.

Related MCP server: LTspice MCP

Supported simulators

Simulator

SPICE_SIMULATOR value

Install

ngspice

ngspice (default)

sudo apt install ngspice

LTspice

ltspice

Download from Analog Devices

Qspice

qspice

Download from Qorvo

Set the SPICE_PATH environment variable to override the simulator binary location.

Setup

Copy .mcp.json.template to .mcp.json in your project directory and adjust SPICE_SIMULATOR if needed:

{
  "mcpServers": {
    "spicelib": {
      "type": "stdio",
      "command": "uvx",
      "args": ["spicelib-mcp"],
      "env": {
        "SPICE_SIMULATOR": "ngspice"
      }
    }
  }
}

Requirements

  • Python 3.10+

  • A supported simulator installed and on PATH

License

GPL-3.0-or-later

Available Tools

4 tools
run_ac_analysisA

Run an AC frequency sweep on a SPICE netlist.

Injects a .ac dec <points_per_decade> <start_freq> <stop_freq> command into a temporary copy of the netlist (does not modify the original).

Results are saved to a .npz file next to the netlist. Load with: data = np.load('netlist_ac.npz') freq = data['frequency_hz'] # traces are stored as complex values, e.g. data['v(out)']

For sweeping a component value or parameter across multiple AC runs in parallel, use run_sweep(analysis_cmd=".ac dec N fstart fstop") instead.

Args: netlist_path: Absolute path to the netlist file (.net, .sp, .cir, etc.) start_freq: Start frequency with optional SPICE suffix, e.g. "1", "1k", "100" stop_freq: Stop frequency, e.g. "1G", "100meg" points_per_decade: Points per decade (default 20)

Returns: JSON summary with data_file path, frequency range, and trace names.

Inputs: netlist_path (required), start_freq (required), stop_freq (required), points_per_decade. Returns: result

run_dc_opA

Run a DC operating point analysis on a SPICE netlist.

Injects a .op command into a temporary copy of the netlist (does not modify the original file). DC op results are small enough to return directly.

Args: netlist_path: Absolute path to the netlist file

Returns: JSON string with keys: type — "dc_op" nodes — dict of node_name → voltage (V) currents — dict of device_name → current (A)

Inputs: netlist_path (required). Returns: result

run_sweepA

Run multiple SPICE simulations in parallel, sweeping component values.

Uses spicelib's SimRunner to execute runs concurrently. Each run gets its own .npz file. Use this instead of calling run_ac_analysis / run_transient in a loop — it's significantly faster for multi-run sweeps.

Example — sweep capacitor C1 across three values with a transient analysis: run_sweep( netlist_path="/path/to/filter.cir", analysis_cmd=".tran 1n 1u", runs=[{"C1": "1p"}, {"C1": "10p"}, {"C1": "100p"}], ) Output files: filter_sweep_001.npz, filter_sweep_002.npz, filter_sweep_003.npz

Each dict in runs maps component reference designators or SPICE parameter names to their new values (strings with optional SPICE suffix, e.g. "4.7k").

Args: netlist_path: Absolute path to the netlist file analysis_cmd: Full SPICE analysis line, e.g. ".tran 1n 1u" or ".ac dec 20 1 1G" runs: List of dicts, one per simulation. Keys are component names or parameter names; values are the new value strings. parallel: Maximum number of simultaneous ngspice processes (default 4)

Returns: JSON list of per-run summaries, each with keys: run — 1-indexed run number data_file — path to the .npz file values — the component/parameter values used for this run traces — list of trace names in the .npz

Inputs: netlist_path (required), analysis_cmd (required), runs (required), parallel. Returns: result

run_transientA

Run a transient simulation on a SPICE netlist.

Injects a .tran <step_time> <stop_time> <start_time> command into a temporary copy of the netlist (does not modify the original).

Results are saved to a .npz file next to the netlist. Load with: data = np.load('netlist_tran.npz') time = data['time_s'] vout = data['v(out)']

For sweeping a component value or parameter across multiple transient runs in parallel, use run_sweep(analysis_cmd=".tran step stop") instead.

Args: netlist_path: Absolute path to the netlist file step_time: Time step with SPICE suffix, e.g. "1n", "10u", "1m" stop_time: Simulation end time, e.g. "1m", "100u" start_time: Time at which to start saving data (default "0")

Returns: JSON summary with data_file path, time range, and trace names.

Inputs: netlist_path (required), step_time (required), stop_time (required), start_time. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.5/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool targets a distinct SPICE analysis type (AC, DC op, transient) or a sweep that combines them. The descriptions clearly state when to use run_sweep instead of individual calls, eliminating ambiguity.

    Naming Consistency5/5

    All tools follow the consistent 'run_<analysis>' pattern with snake_case. The naming is predictable and makes the purpose immediately clear.

    Tool Count5/5

    With 4 tools covering the core SPICE analyses and a parallel sweep utility, the count is well-scoped for a simulation-focused server. No extraneous or redundant tools.

    Completeness4/5

    The core transient, AC, and DC op analyses are covered, and the sweep tool fills the gap for multi-run parameter sweeps. Missing a dedicated DC sweep tool, but the sweep tool can handle that via analysis_cmd, so the gap is minor.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      C
      quality
      C
      maintenance
      MCP server for automating LTspice on macOS, enabling simulation, schematic generation, data extraction, verification, and rendering via natural language or agents.
      71
      22
      -
    • A
      license
      A
      quality
      B
      maintenance
      An MCP server that connects LLM assistants to real circuit simulation: LTspice and ngspice, plus direct editing of LTspice .asc schematics. Simulation results come back as structured numbers so the assistant can design, verify, and iterate on circuits.
      48
      868 PyPI
      81
      GPL 3.0
    • F
      license
      Not graded
      quality
      D
      maintenance
      Provides circuit simulation capabilities via MCP, enabling creation, simulation (DC, AC, transient), and analysis of electronic circuits using PySpice.
      1
      -