microcap-mcp
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., "@microcap-mcp Simulate transient response of RC lowpass filter with 1kHz cutoff"
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.
microcap-mcp
ИИ-агент строит, считает и рисует аналоговые схемы в Micro-Cap 12.
Русский · English
Схему нарисовал сам сервер (1:1 с Micro-Cap, с рабочей точкой) · график посчитал Micro-Cap
Работает через штатный batch-режим Micro-Cap (MC12 @batch.bat) — headless, без GUI-автоматизации и без модификации программы.
Что делает
🔧 Строит схемы с нуля — 8 генераторов каскадов (общий эмиттер, повторитель, MOSFET, дифпара, токовое зеркало, каскод, ОУ, RLC) с авто-смещением. Каждый сверен с теорией: усиление, рабочая точка, резонанс.
📊 Считает — transient / AC / DC / искажения / устойчивость, рабочая точка, свипы. Комплексный вывод как {re, im}, диагностика солвера.
🎨 Рисует — сам рендерит .CIR в SVG 1:1 с Micro-Cap: родные символы из библиотеки MC, сетка, все повороты и отражения, раскладка подписей. Накладывает напряжения рабочей точки. /IC в batch-режиме MC этого не умеет — а сервер умеет.
Related MCP server: ltspice-mcp
Установка
Нужен Windows, Python 3.11+ и установленный Micro-Cap 12 (фримварь). Клонировать ничего не надо — укажи в MCP-клиенте, и uvx сам скачает пакет:
{
"mcpServers": {
"microcap": {
"command": "uvx",
"args": ["microcap-mcp"],
"env": { "MICROCAP_HOME": "E:/Tools/MC12" } // только если автопоиск не нашёл
}
}
}Драйвер сам ищет Micro-Cap в типовых местах; MICROCAP_HOME укажи на папку с mc12_64.exe, если он лежит нестандартно.
git clone https://github.com/monoxide-xen/microcap-mcp
cd microcap-mcp
uv sync{
"mcpServers": {
"microcap": {
"command": "uv",
"args": ["--directory", "C:/path/to/microcap-mcp", "run", "microcap-mcp"]
}
}
}Инструменты
Инструмент | Что делает |
| посчитать SPICE-нетлист, свип по |
| каскады на транзисторе / MOSFET, авто-смещение под середину питания |
| дифпара, зеркало, каскод, усилитель на ОУ |
| источник + R/C/L последовательно и параллельно (RC/RL/RLC, делители, контуры) |
| посчитать / построить график произвольной |
| нарисовать |
| ~490 эталонных схем из поставки MC: поиск, исходник, анализы |
Плюс ресурсы microcap://guide и microcap://domains и промпт analyse_circuit — чтобы агент не дал MC молча соврать.
Тесты
uv run pytest # 138 юнит без Micro-Cap; +30 интеграционных против физики (нужен MC)Без Micro-Cap интеграционные пропускаются — CI остаётся зелёным.
Как это устроено
Особенности Micro-Cap — поведение, которого нет в мануале (а местами оно мануалу противоречит).
eval/harness.py— прогон всех ~490 схем поставки с раскладкой провалов по причинам (сейчас 88% отвечают).
Лицензия
MIT на код репозитория. Micro-Cap 12 принадлежит Spectrum Software — не включён, не перезалит, не модифицирован: используется его документированный CLI.
Available Tools
20 toolsannotate_schematicA
Render the schematic as SVG with each labelled node's DC operating-point voltage written beside it — a marked-up drawing you can read the bias off.
Runs the DC operating point (a short transient at t=0), reads the voltage at
every [Grid Text] node the schematic labels (IN, OUT, OUTP…),
and overlays them on the drawing. Best on the generated stages and other
circuits that hold a stable operating point.
Inputs: cir (required).
describe_exampleA
Report what a reference circuit is set up to do, without running it.
Returns the analyses it defines and the expressions it plots. Read this
before simulate_example so you ask for an analysis that exists.
Inputs: name (required). Returns: object
draw_schematicA
Render a .CIR schematic as an SVG picture — the drawing, not a plot.
Micro-Cap's own schematic-image command does not work in batch, so this
draws the circuit directly from the .CIR (the same placements, wires and
node labels), giving recognisable symbols for R/C/L, sources, ground,
BJT/MOSFET and op-amp. Use it to see a circuit — a reference you fetched or
one drawn by the generators — without opening Micro-Cap. No simulation runs.
Returns SVG (vector, self-contained); save it as .svg or embed it.
Inputs: cir (required).
generate_amplifierA
Draw a .CIR op-amp amplifier with a near-ideal op-amp.
inverting gives gain -Rf/Rin; non-inverting gives 1 + Rf/Rg.
The feedback resistor is sized from gain and rin. Produces a drawn
schematic; feed it to simulate_schematic or hand the .CIR to the user.
Args:
gain: magnitude of the closed-loop gain (> 0; non-inverting needs >= 1).
kind: inverting or non-inverting.
rin: input resistor (inverting) or gain resistor to ground
(non-inverting); sets the feedback resistor from the gain.
source: source VALUE, Micro-Cap syntax ("DC=0 AC=1" for AC).
analysis: AC, Transient, or DC.
output_node: label for the output node.
Returns the .CIR text.
Inputs: gain (required), rin, kind, source, +2 more. Returns: object
generate_cascodeA
Draw a .CIR cascode amplifier (common-emitter under common-base).
A common-emitter transistor stacked under a common-base transistor. The
midband gain is the common-emitter's, -Rc/(Re+re'), but the upper
transistor shields the lower one's collector from the output swing, so the
Miller capacitance nearly vanishes and both output impedance and bandwidth
rise — the reason to use a cascode over a plain common-emitter. All bias
points are computed for a mid-supply output with active-region headroom, and
both bases are driven by sources carrying their DC bias directly (no coupling
caps or dividers). Rc must exceed Re.
Args:
rc, re: collector load and emitter degeneration; their ratio is the gain.
vcc: supply voltage.
source: AC drive VALUE added to the lower base's bias ("AC=1").
analysis: AC, Transient, or DC.
output_node: label for the output node (the upper collector).
Returns the .CIR text.
Inputs: rc, re, vcc, source, +2 more. Returns: object
generate_current_mirrorA
Draw a .CIR BJT current mirror.
A diode-connected NPN sets a reference current Iref = (Vcc - Vbe)/Rref;
a matched NPN copies it into Rload. The mirrored current is
Iout ≈ Iref (a few % high from the Early effect). It is a DC bias block,
so run a transient/DC and read V(OUTC): the mirrored current is
(Vcc - V(OUTC)) / Rload.
Args:
rref: reference-leg resistor; sets Iref.
rload: load on the output transistor's collector.
vcc: supply voltage.
analysis: transient or dc (it is a bias block, not an AC stage).
output_node: label for the output collector node.
Returns the .CIR text.
Inputs: vcc, rref, rload, analysis, +1 more. Returns: object
generate_differential_pairA
Draw a .CIR BJT long-tailed (differential) pair.
Two matched NPN sharing an emitter tail resistor to ground, each collector
loaded by Rc. Single-ended drive on one base; the collectors OUTP
and OUTN swing in antiphase, each with magnitude Rc/(2*re'). The
common-mode base voltage and tail resistor are auto-sized to bias both
collectors near mid-supply, so the pair is balanced and in the active region
— the differential gain then works out near Vcc/(4*Vt), set by the
supply rather than by Rc. Both collectors are labelled; the default
trace plots OUTP (edit it to plot OUTN).
Args:
rc: collector load on each side.
rt: emitter tail resistor; leave unset to auto-size for mid-supply.
vcc: supply voltage.
vb: common-mode base voltage; leave unset for vcc/2.
source: AC drive VALUE for the input base ("AC=1").
analysis: AC, Transient, or DC.
output_node: which collector the default trace plots (OUTP/OUTN).
Returns the .CIR text.
Inputs: rc, rt, vb, vcc, +3 more. Returns: object
generate_emitter_followerA
Draw a .CIR emitter follower (common-collector BJT buffer).
Collector straight to the supply, output at the emitter. Voltage gain is
just under 1 (Re/(Re+re')); the value is current gain and low output
impedance — a buffer that drives a heavy load without loading the source.
The divider auto-biases the emitter at mid-supply so it can swing both ways.
Feed the result to simulate_schematic.
Args:
re: emitter resistor (sets the bias current Vcc/2 / Re).
r1, r2: base bias divider; leave unset to auto-bias, or give both.
vcc: supply voltage.
cin: input coupling capacitor.
source: input source VALUE, Micro-Cap syntax.
analysis: AC, Transient, or DC.
output_node: label for the emitter output node.
Returns the .CIR text.
Inputs: r1, r2, re, cin, +4 more. Returns: object
generate_mosfet_amplifierA
Draw a .CIR common-source MOSFET gain stage (NMOS primitive).
Gate divider bias, source degeneration, AC-coupled input, body tied to
source. Midband gain is -gm*Rd/(1+gm*Rs) — roughly -Rd/Rs when
gm*Rs is large. By default the gate divider is computed from the model's
VTO/KP to bias the drain at mid-supply, so the device stays in
saturation and actually amplifies (a mis-biased MOSFET drops out of
saturation and the gain collapses). Feed the result to simulate_schematic.
Args:
rd, rs: drain and (degeneration) source resistors.
r1, r2: gate bias divider; leave unset to auto-bias for a mid-supply
drain, or give both to override.
vdd: supply voltage.
cin: input coupling capacitor.
source: input source VALUE, Micro-Cap syntax ("AC=1" for AC gain).
analysis: AC, Transient, or DC.
output_node: label for the drain output node.
Returns the .CIR text.
Inputs: r1, r2, rd, rs, +5 more. Returns: object
generate_schematicA
Draw a .CIR schematic: a voltage source driving parts in series to
ground, optionally with parallel branches from the output node to ground.
Unlike a netlist, this produces a drawn schematic you can open in
Micro-Cap. Feed the result to simulate_schematic to run it, or hand the
.CIR text to the user.
Bounded on purpose — a source, a series chain of two-terminal passives (R, C, L), and optional parallel shunt branches. That covers RC/RL/RLC, dividers, and resonant tanks. Every pin position is taken from Micro-Cap's own component library, so the drawn circuit is electrically what you asked for. For an op-amp gain stage use generate_amplifier; for a common-emitter BJT stage use generate_transistor_amplifier.
Args:
parts: ordered "KIND=VALUE" strings in series, KIND in R/C/L, e.g.
["R=1K", "C=159.155N"] for an RC low-pass. With shunt, the
whole chain feeds the output node.
source: the source's VALUE in Micro-Cap syntax — "DC=0 AC=1" for an
AC probe, or a PULSE ... line for transient.
analysis: AC, Transient, or DC.
output_node: label for the output node.
shunt: extra "KIND=VALUE" strings hung in parallel from the output
node to ground, e.g. ["L=1M", "C=1U"] on a series R for a tank.
Returns the .CIR text and the ordered part references.
Inputs: parts (required), shunt, source, analysis, +1 more. Returns: object
generate_transistor_amplifierA
Draw a .CIR common-emitter BJT gain stage (NPN primitive).
Divider bias, unbypassed emitter degeneration, AC-coupled input. The
midband gain magnitude is Rc/(Re+re') — roughly Rc/Re when Re is
well above the intrinsic re'. By default the R1/R2 divider is
computed to bias the collector at mid-supply, so the stage stays in the
active region for any Rc/Re and the gain is real (a fixed divider
would silently saturate as Rc grows). Rc must exceed Re. Feed
the result to simulate_schematic.
Args:
rc, re: collector and (unbypassed) emitter resistors; their ratio is the
midband gain. Rc must exceed Re.
r1, r2: base bias divider (vcc to base, base to ground); leave unset
to auto-bias for a mid-supply collector, or give both to override.
vcc: supply voltage.
cin: input coupling capacitor.
source: input source VALUE, Micro-Cap syntax ("AC=1" for AC gain).
analysis: AC, Transient, or DC.
output_node: label for the collector output node.
Returns the .CIR text.
Inputs: r1, r2, rc, re, +5 more. Returns: object
get_exampleA
Fetch one reference circuit's source text.
.CKT files are plain SPICE and can be fed straight to simulate.
.CIR files are Micro-Cap schematics: text, but carrying component
coordinates and analysis settings as well as the netlist.
Inputs: name (required). Returns: object
list_domainsA
List the circuit-design domains Micro-Cap ships reference circuits for.
Returns domain name -> circuit count. Start here before designing from scratch: a working reference beats an invented topology.
Inputs: no named parameters. Returns: object
plotA
Run a netlist and return Micro-Cap's own rendered plot as a JPEG image.
Use this to look at a waveform. For numbers to reason about, use
simulate — reading values off a picture is guesswork.
Inputs: netlist (required), defines, analysis.
plot_schematicA
Run a .CIR schematic and return Micro-Cap's rendered plot as a JPEG.
The picture companion to simulate_schematic: use it to look at the
response of a schematic — a reference circuit you edited, or one drawn by
generate_transistor_amplifier / generate_differential_pair / the
other generators. For numbers to reason about, use simulate_schematic.
Inputs: cir (required), points, analysis.
search_examplesB
Search the ~470 reference circuits by name or domain.
Inputs: query (required), limit. Returns: result
simulateA
Run a SPICE netlist through Micro-Cap and return the waveform data.
The netlist is plain SPICE. Rules that will bite you if ignored:
Node
0is ground and must exist. Every node needs a DC path to it, or the run fails to converge.A
.PRINTline is mandatory — it names the outputs you want. Without it Micro-Cap emits only an operating-point dump and no waveforms.Match the analysis to the question:
acfor frequency response and gain/phase,transientfor time-domain behaviour and start-up,dcfor bias points and transfer curves.Set the time span from the circuit's own constants. A
.TRANrunning for 10 s on a circuit with microsecond edges returns a flat line.Active parts need their supplies wired; an op-amp without rails does nothing.
Args:
netlist: full SPICE deck, first line is the title, ending in .END.
analysis: transient | ac | dc | harmonic_distortion |
intermodulation_distortion | dynamic_ac | dynamic_dc | stability.
defines: values for .DEFINE symbols in the deck, e.g. {"R": "1K"}.
max_points: cap on returned samples; the run itself is unaffected.
Returns: Columns, units, sampled data, solver statistics, and any warnings.
Inputs: netlist (required), defines, analysis, max_points. Returns: object
simulate_exampleA
Run one of Micro-Cap's own reference circuits and return its data.
These are worked designs by the tool's authors, so this is the cheapest way
to get a trustworthy baseline before modifying anything. Use
describe_example first to see which analyses a circuit supports and
what it plots — asking for an analysis it was not built for returns nothing
useful.
Numeric export is enabled automatically; the shipped circuits have it off.
Args:
name: reference circuit name.
analysis: which analysis to run.
max_points: cap on returned samples.
points: resolution to compute at. The reference circuits carry their
own display setting, often tiny or 0 — which exports a single
useless row and makes an oscillator look dead. Overriding it is what
makes the library usable as data; only lower it if you specifically
want the circuit's own resolution.
Inputs: name (required), points, analysis, max_points. Returns: object
simulate_schematicA
Run a Micro-Cap .CIR schematic and return its data.
Use this to adapt a reference: fetch one with get_example (its
format will be microcap_schematic), change a component value or two
in the text, and run the modified schematic here. simulate is for plain
SPICE netlists; this is for the coordinate-carrying .CIR format.
Numeric export is switched on automatically, NPts is overridden to
points (the shipped setting is often 0, which exports one row), and
the symbolic transient bound TMIN is repaired for batch mode — all the
things that make a stock schematic yield no data otherwise.
Args:
schematic: full .CIR text.
analysis: which of the circuit's analyses to run.
defines: values for any .DEFINE symbols the schematic uses.
points: resolution to compute at; see simulate_example.
max_points: cap on returned samples.
Inputs: schematic (required), points, defines, analysis, +1 more. Returns: object
sweepA
Run one netlist repeatedly, varying a .DEFINE parameter.
The deck must declare the symbol, e.g. .DEFINE RLOAD 1K, and use it in
a component value. Each value is a separate Micro-Cap run.
Args:
netlist: SPICE deck containing a .DEFINE for parameter.
parameter: the symbol to vary.
values: values to substitute, e.g. ["1K", "10K", "100K"].
analysis: analysis to run for every value.
max_points: cap on returned samples per run.
Inputs: values (required), netlist (required), parameter (required), analysis, +1 more. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 20 tools
Several tools have overlapping names and purposes: simulate, simulate_schematic, and simulate_example are all 'run a simulation' tools with different input formats, and plot and plot_schematic similarly differ only by input type. The descriptions clearly distinguish them, but an agent relying on names alone could easily misselect.
The naming pattern is predominantly verb_noun with underscores (generate_amplifier, simulate_schematic, describe_example), which is consistent. A few single-word verbs (simulate, sweep, plot) deviate slightly but remain predictable and readable.
At 20 tools, the server is slightly over the ideal 3-15 range but each tool serves a distinct purpose in the circuit simulation workflow. The count is justifiable given the breadth of generation, simulation, visualization, and example-library access, though a few generators could potentially be consolidated.
The tool surface covers the core lifecycle: generate circuits, simulate via netlist or schematic, plot results, and access reference examples. Minor gaps exist, such as no explicit schematic editing tool (text editing is expected instead) and no Monte Carlo or corner analysis, but the main workflows are well supported.
Related MCP Connectors
MCP server for building and testing AI agents with multi-model experimentation and insights.
MCP server for AI dialogue using various LLM models via AceDataCloud
MCP server exposing the Backtest360 engine API as tools for AI agents.
Trading MCP server for AI agents, with live market data, account reads and controlled execution.
Related MCP Servers
- AlicenseBqualityDmaintenanceAI-powered circuit design through simulation — an MCP server that gives language models direct access to SPICE circuit simulation via ngspice, enabling natural language circuit description and automated netlist generation, simulation, measurement, and spec verification.28107 PyPI47GPL 3.0
- FlicenseCqualityCmaintenanceMCP server for automating LTspice on macOS, enabling simulation, schematic generation, data extraction, verification, and rendering via natural language or agents.7122-
- AlicenseNot gradedqualityBmaintenanceThis MCP server enables agents to control LTspice on macOS for running simulations, generating schematics, extracting data, and automating verification workflows.MIT
- AlicenseAqualityBmaintenanceAn 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.48868 PyPI81GPL 3.0