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., "@circuitarium-mcp Run electrical rule checks on the CRUMB project and generate a BOM."
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.
Local electronics tools for MCP-capable assistants, with support for CRUMB save files and Logisim-evolution projects.
Circuitarium lets an MCP host inspect circuit files without pretending that static analysis is a running simulator. It can trace CRUMB nets, build BOMs, run electrical rule checks, analyze Logisim projects, and optionally ask Logisim-evolution to produce truth tables or run test vectors.
Release 0.3.1 includes 22 tools, nine read-only Resources, and four workflow
Prompts. The server is model-neutral: Codex, Claude Code/Desktop, VS Code,
LM Studio, Jan, or another local MCP host supplies the model and launches
Circuitarium over stdio. Circuitarium itself needs no OpenAI, Anthropic, or
local-model API key.
You need:
Node.js 22 or newer;
an MCP client that can launch a local stdio server; and
a dedicated folder containing only the circuit files you want the client to access.
Windows, macOS, and Linux are covered by CI. CRUMB analysis and the static Logisim tools need neither simulator nor Java.
To install and start the local stdio server from a terminal:
npx -y circuitarium-mcp@0.3.1In an interactive terminal this prints a short status panel. In normal use, your MCP client runs the same command and keeps the server connected over stdio.
Most clients ask for the same command, arguments, and environment variable:
Command: npx
Arguments: -y circuitarium-mcp@0.3.1
Environment:
CIRCUITARIUM_MCP_ROOT=/absolute/path/to/circuit-workspaceThe first launch can take longer while npm downloads the package. Later starts normally use npm's cache.
Replace the path with the smallest folder that should contain the .cru,
.circ, .vec, and .txt files available to the model. Do not point it at
your home directory or a drive root.
Copyable configurations are available for:
See the client setup guide for platform-specific examples and local-model guidance. The starter workspace recipe copies one synthetic CRUMB fixture plus the full-adder Logisim project into an isolated folder, so a first test does not require access to the rest of a source checkout. Restart or reload the client after adding the server.
Claude Desktop users can instead download
circuitarium-mcp-0.3.1.mcpb
directly and open it as a local bundle. The installer asks for a circuit
workspace and offers optional Logisim JAR and Java settings. The
release page
also includes checksums and the matching npm tarball.
Paste this into the connected client:
Use Circuitarium. Call
electronics_capabilities. Create the built-inbreadboard-ledfixture asdemo/first-led.cruwithout overwriting anything, analyze it, and runcrumb_check_design. Explain which findings are static inferences rather than simulation results.
Your client may ask you to approve creation of the synthetic file. The tool refuses to overwrite an existing path; on a repeat run, analyze the existing file or choose a new name.
A healthy connection creates a small synthetic file, recognizes its breadboard and LED, and reports two floating-terminal warnings. The fixture is independently authored and safe to redistribute; it is not a CRUMB asset.
These commands verify the published launcher without opening a server session:
npx -y circuitarium-mcp@0.3.1 --help
npx -y circuitarium-mcp@0.3.1 --version
npx -y circuitarium-mcp@0.3.1 doctor --smokeIf the bare npx -y circuitarium-mcp@0.3.1 command appears to wait, the server
is working as designed: it is waiting for an MCP host on stdin. Press Ctrl+C
and let the host launch the command itself.
Watch the terminal demo to see the source-checkout command runner inspect a synthetic fixture, build its BOM, and export its netlist. The public package exposes the MCP server, help, version, and doctor commands shown above.
Watch the MCP Inspector demo to see the server running in an independent MCP client, or read the reproducible 22-tool host verification and its sanitized evidence report.
Related MCP server: kicad-forge
Area | Current capabilities |
General electronics | Discover available backends, validate portable experiment descriptions, and plan what evidence is needed to verify circuit claims |
CRUMBLE / CRUMB | Discover, inspect, validate, and compare Unity-era |
Logisim-evolution | Discover and inspect |
Cross-model work | Preserve workspace-relative project references and raw-byte SHA-256 digests so another model can confirm it is reading the same artifact |
CRUMBLE—Circuit Representation & Universal Model Bridge for Laboratory Electronics—is the unofficial CRUMB-specific integration inside Circuitarium. The simulator-neutral tools keep CRUMB-specific assumptions out of future adapters.
The current surface breaks down into three neutral electronics_* tools,
thirteen crumb_* tools, and six logisim_* tools. All tool results use the
electronics.mcp/0.2 result contract, independent of the package version.
Operation | Simulator or Java required? | What it establishes |
All | No | Facts inferred from a saved Unity-era |
| No | Saved |
| Java 21 and a user-supplied Logisim 4.1.0 all-JAR | The selected project and circuit loaded in one bounded subprocess |
| Java 21 and a user-supplied Logisim 4.1.0 all-JAR | Bounded behavioral evidence for the exact project digest; not a live GUI session |
Resources and Prompts | No | Guidance and workflow templates only; they do not read a project or execute a simulator |
Target | Status | Boundary |
CRUMB 1.3.5, Unity-era | Experimental support | Static, version-pinned file analysis under |
Godot-era CRUMB 2.x | Not supported | Requires a separate evidence profile and controlled-save corpus |
Live CRUMB control or simulation | Not implemented | No run, pause, step, signal-read, or GUI bridge |
Arbitrary CRUMB editing | Not implemented | The only public write tool creates one of five fixed synthetic fixtures |
Logisim-evolution 4.1.0 static analysis | Experimental support | Clean-room |
Logisim-evolution 4.1.0 execution | Optional | Bounded, one-shot JAR subprocesses; no live GUI session |
CRUMB electrical rule checks operate on inferred net connectivity and saved component values. They can find supply shorts, an LED placed directly across both rails of one recognized supply, a directly connected resistor whose saved values imply excessive dissipation, and selected floating terminals or IC power pins. They do not trace series paths, prove polarity, calculate general circuit behavior, or certify a physical build.
Circuitarium is experimental educational software, not a safety certification tool. Do not rely on its output alone for mains voltage, battery charging, medical, automotive, life-safety, high-energy, or regulatory work. Use authoritative component documentation, appropriate measurements, and qualified engineering review wherever consequences matter.
File parsing happens locally, but the MCP host may send returned data to a cloud model. Embedded firmware source is omitted by default, and binary EEPROM contents are never returned, but you should still use a narrow workspace with no secrets. Read the security policy for the full trust model.
The three static Logisim tools work without Java. To enable component statistics, truth tables, and test vectors:
Install Java 21 or newer.
Download the official
logisim-evolution-4.1.0-all.jar.
Add the JAR path to the MCP host configuration.
CIRCUITARIUM_LOGISIM_JAR=/absolute/path/to/logisim-evolution-4.1.0-all.jar
CIRCUITARIUM_JAVA=javaCIRCUITARIUM_JAVA is optional when java already resolves to Java 21 or
newer. On a headless Linux machine, test-vector execution also needs a working
X11 display; Xvfb is sufficient.
Use only a trusted JAR. Circuitarium verifies that it self-reports version 4.1.0, but that response does not authenticate the file or its publisher. Runtime tools may update Logisim's per-user Java preferences and are not an operating-system sandbox.
The repository includes an independently authored full-adder project and vector for a first runtime test. Circuitarium does not bundle or download Logisim-evolution. See the Logisim adapter guide for setup, Linux notes, evidence levels, and subprocess safeguards.
Call electronics_capabilities.
Call crumb_list_projects if the path is not known. Projects include a
digest when requested and within the documented file and aggregate bounds;
otherwise the result explains why it was omitted.
Call crumb_analyze_design with view: "summary".
Use crumb_export_netlist, crumb_trace_net, crumb_check_design, or
crumb_bom only as the question requires.
Static findings remain file-format inference. Circuitarium does not observe whether CRUMB is running or how the circuit behaves.
Call logisim_list_projects, then logisim_analyze_design.
Treat logisim_export_netlist as partial static evidence and review its
conversion-loss markers.
If the optional runtime is available, select an exact circuit and call
logisim_truth_table or logisim_run_test_vector.
Use electronics_plan_verification when a claim needs several kinds of
evidence or an explicit list of remaining gaps.
Keep projectRef, projectDigest, backendId, adapterVersion, and
compatibilityProfile with the handoff. The receiving model should pass the
recorded digest as expectedProjectDigest on its first read. If the bytes have
changed, Circuitarium returns PROJECT_STATE_CONFLICT instead of silently
reusing stale conclusions.
See the cross-model handoff example.
Tool | Purpose |
| Report callable backends, limitations, and recommended workflows |
| Validate a portable experiment description without claiming it ran |
| Build an evidence plan for explicit circuit claims |
Tool | Purpose |
| Return the version-pinned CRUMB component catalog |
| Read bounded semantic views of a |
| Compare two digest-guarded saves without modifying either |
| Return a compact format-level inventory |
| Check XML structure and known invariants |
| Create one fixed synthetic fixture without overwriting |
| Discover workspace |
| Fetch one component and optional bounded source window |
| Group recognized components into a bill of materials |
| Query the version-pinned IC package and pin registry |
| Export static, jumper-collapsed inferred nets |
| Produce a paged conductive witness for one terminal |
| Run scoped static electrical rule checks |
Tool | Purpose |
| Discover workspace |
| Inspect static project, circuit, component, Pin, Clock, and wire structure |
| Export an explicitly partial coordinate-based neutral netlist |
| Ask the configured JAR to load and inventory a circuit |
| Evaluate a bounded combinational truth table |
| Run a workspace-contained vector and return structured failures |
The nine Resources provide capability metadata, compatibility profiles, synthetic-example metadata, electrical-review and digital-testing guidance, a neutral component-profile schema, and version-pinned component catalogs. The four user-invoked Prompts are:
review-circuit-design
compare-crumb-designs
verify-logisim-design
handoff-circuit-project
Resources and Prompts do not read a project or run a simulator by themselves. Clients that do not expose those MCP features can still use all 22 tools. The contract reference documents exact resource URIs, Prompt arguments, result envelopes, errors, output bounds, and tool schemas.
Circuitarium is still early, and small, well-evidenced improvements are useful. Good starting points include client setup fixes, synthetic test circuits, source-cited component profiles, parser edge cases, controlled compatibility observations, and scoped roadmap work.
Before changing an adapter or public contract, read the contribution guide, architecture, and provenance policy. Questions and early design ideas belong in Discussions; reproducible bugs, scoped feature requests, and controlled interoperability evidence belong in the issue tracker.
git clone https://github.com/Craftiee/circuitarium-mcp.git
cd circuitarium-mcp
npm ci
npm run lint
npm run checkUse npm run test:coverage for behavioral changes,
npm run package:check for package or launcher changes, and
npm run logisim:e2e for Logisim runtime changes. The Logisim E2E test
requires Java 21 and a separately supplied 4.1.0 all-JAR.
Fixtures must be small, synthetic, independently authored, and safe to redistribute. Do not submit private circuits, firmware, simulator binaries, extracted assets, or third-party circuit designs. Build a minimal synthetic reproduction instead. Report vulnerabilities through GitHub private vulnerability reporting, not a public issue. All contributors must follow the Code of Conduct.
Document | Use it for |
Host-specific installation, subscriptions/APIs, and local models | |
Current CRUMB scope and evidence profiles | |
JAR setup, runtime evidence, safety, and Linux notes | |
Exact tools, Resources, Prompts, envelopes, and limits | |
Adapter boundaries and simulator-neutral design | |
Evidence categories and redistribution rules | |
Shipped milestones and planned work | |
Release history |
Circuitarium runs locally over stdio, has no telemetry, and does not send circuit files to a maintainer-controlled service. The MCP host and model provider you choose may still process tool arguments and results under their own terms. Read the complete Privacy Policy before granting a host access to a workspace.
Circuitarium MCP is experimental community software released under the
Apache License 2.0. Version 0.3.1 is available on
npm and as a
GitHub release,
and its metadata is listed in the
official MCP Registry.
For help, see SUPPORT.md. Research and teaching users can cite
the project with CITATION.cff.
Circuitarium MCP, CRUMBLE, and the Logisim-evolution adapter are independent, unofficial interoperability work. They are not affiliated with, endorsed by, or sponsored by CRUMB, Logisim-evolution, or their developers. This repository contains no CRUMB or Logisim-evolution simulator code, binaries, extracted assets, logos, or bundled third-party circuit designs. Product names and marks belong to their respective owners.
Recognizes version-pinned component parameters, terminal attachments, and inferred connection groups. Summary is the bounded default; components and connections use opaque cursors. Embedded firmware and geometry require explicit opt-in.
Inputs: path (required), view, limit, cursor, +4 more. Returns: ok, data, error, context, +4 more
Groups recognized components by kind and decoded part values into quantities. State values such as switch positions are excluded from part identity; unknown and schema-mismatched components stay visible as their own lines.
Inputs: path (required), limit, expectedProjectDigest. Returns: ok, data, error, context, +4 more
Lints the inferred netlist: supply shorts, LEDs directly across the rails, shorted two-terminal parts, floating IC power pins, resistor power ratings, and floating terminals. Findings carry evidence confidence and rule basis; a rule violation returns ok=true with data.valid=false.
Inputs: path (required), topologyMode, applySwitchStates, expectedProjectDigest. Returns: ok, data, error, context, +4 more
Read-only, GUID-matched comparison of a baseline and candidate .cru under crumb.unity/1.3.5. Distinguishes exact bytes, modeled equivalence, root changes, component changes, and unverified payload signatures without returning raw XML, firmware, EEPROM bytes, or thumbnails.
Inputs: baselinePath (required), candidatePath (required), view, limit, +6 more. Returns: ok, data, error, context, +4 more
Returns the version-pinned CRUMB tool-ID catalog, payload signatures, typed parameters, terminal labels, confidence values, and their machine-readable evidence vocabulary. Optionally filter by toolId.
Inputs: toolId. Returns: ok, data, error, context, +4 more
Collapses jumper wires out of the inferred connection graph and returns paged electrical nets with component terminals, VCC/GND names inferred from DC supply terminals, and optional saved-switch-state merges. Provenance and confidence are explicit.
Inputs: path (required), limit, cursor, topologyMode, +2 more. Returns: ok, data, error, context, +4 more
Creates one known fixture without overwriting. Provide outputPath for a file artifact; raw XML is returned only when includeXml=true.
Inputs: kind (required), name, includeXml, outputPath. Returns: ok, data, error, context, +4 more
Returns a single component by id with parameters, terminals, geometry, its inferred connection groups, and windowed access to embedded firmware source past the analyze cap. The read-back companion for iterative work on one part.
Inputs: path (required), componentId (required), sourceOffset, topologyMode, +3 more. Returns: ok, data, error, context, +4 more
Queries the version-pinned tool-5 IC registry by prefabId or by a label/package substring (for example "74HC138"). Returns package labels, ordered pin names, and explicit unresolved pins.
Inputs: query, prefabId. Returns: ok, data, error, context, +4 more
Returns a bounded save summary and tool-ID counts. Use crumb_analyze_design for paginated semantic component and connection details.
Inputs: path (required), expectedProjectDigest. Returns: ok, data, error, context, +4 more
Enumerates .cru files under the workspace root (or one subdirectory) with size, modification time, and SHA-256 digest, so a model can discover projects without being handed a path.
Inputs: dir, limit, recursive, includeDigests. Returns: ok, data, error, context, +4 more
Selects one component terminal by stable index and returns a paged deterministic connectivity witness with structured attachment, board, jumper, and optional saved-switch provenance. It is static conductive inference, not current flow, path enumeration, or simulation.
Inputs: path (required), componentId (required), terminalIndex (required), limit, +5 more. Returns: ok, data, error, context, +4 more
Performs XML and CRUMB structural checks without launching the game. A bad design returns ok=true and data.valid=false.
Inputs: path (required), expectedProjectDigest. Returns: ok, data, error, context, +4 more
Zero-argument onboarding for model-neutral conventions, callable backends, truthful limitations, and recommended workflows. Call this first when unsure.
Inputs: no named parameters. Returns: ok, data, error, context, +4 more
Builds a deterministic, simulator-neutral plan for explicit claims using bounded caller-reported evidence. It reads no files, runs no tools or simulators, authenticates no receipts, and never certifies physical hardware.
Inputs: claims (required), target (required), evidence, declaredInterface. Returns: ok, data, error, context, +4 more
Checks an arbitrary JSON value against the simulator-neutral circuit, firmware, probe, assertion, and timing contract. Invalid input returns ok=true and data.valid=false with diagnostics.
Inputs: experiment (required). Returns: ok, data, error, context, +4 more
Parses bounded .circ XML into project, circuit, pin, clock, component, and explicit conversion-loss summaries. Static parsing is not simulation evidence.
Inputs: path (required), circuit, expectedProjectDigest. Returns: ok, data, error, context, +4 more
Invokes the separately installed JAR, after it self-reports Logisim-evolution 4.1.0, with --tty stats. Success proves that configured process loaded the staged project, not behavioral simulation or binary authenticity.
Inputs: path (required), limit, circuit, timeoutMs, +1 more. Returns: ok, data, error, context, +4 more
Exports simulator-neutral coordinate-endpoint nets with explicit loss markers. It does not infer unmodeled gate geometry, mid-segment junctions, or behavior.
Inputs: path (required), limit, cursor, circuit, +1 more. Returns: ok, data, error, context, +4 more
Discovers workspace .circ projects with stable raw-byte digests. This is static file discovery and does not launch Logisim.
Inputs: dir, limit, recursive. Returns: ok, data, error, context, +4 more
Invokes the separately installed JAR, after it self-reports Logisim-evolution 4.1.0, with staged snapshots of a workspace-contained project and .vec/.txt file. Assertion failures return ok=true with data.valid=false.
Inputs: path (required), vectorPath (required), circuit, timeoutMs, +3 more. Returns: ok, data, error, context, +4 more
Invokes the separately installed JAR, after it self-reports Logisim-evolution 4.1.0, in CSV/binary table mode after statically bounding declared input width.
Inputs: path (required), limit, circuit, timeoutMs, +2 more. Returns: ok, data, error, context, +4 more
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 22 tools
Every tool has a clearly distinct purpose, reinforced by consistent prefixes (crumb_, logisim_, electronics_) and explicit descriptions. Even closely related tools like crumb_validate_design (structural XML checks) and crumb_check_design (netlist linting) are unambiguously separated. Higher-level analysis tools (inspect vs analyze) are differentiated by summary vs paginated detail.
The naming is mostly consistent with a predictable pattern: prefixes for simulators, and parallel names for shared operations (list_projects, analyze_design, export_netlist). Minor deviations exist, such as noun-phrase names (crumb_bom, crumb_ic_reference, electronics_capabilities) versus verb-first names, but the overall structure remains readable and coherent.
At 22 tools, this exceeds the typical 3-15 range, but the count is justified by covering two distinct simulators (CRUMB and Logisim) plus a generic electronics validation layer. Each tool serves a unique function, and the number feels appropriate for the breadth of the domain.
The tool set provides comprehensive coverage for circuit analysis and validation: project discovery, design analysis, validation, netlist export, tracing, BOM generation, IC lookup, and simulation/test execution. Minor gaps exist, such as lack of design creation/editing tools, but these appear outside the server's stated analytical purpose.
Design perfboard and stripboard layouts, check the wiring and get a link to open them in the editor.
Turn datasheets into source-backed JSON contracts, pinout matrices, and KiCad CAD for AI agents.
Run ERC, DRC, parity, BOM and export checks on KiCad 9 boards before manufacturing.
Design, route, check, review and export manufacturable PCBs from plain language; KiCad import.
