multisim-circuit-workflow

Multisim Circuit Workflow

Use the multisim MCP tools in this order. Do not skip pre-validation.

0. Verify the installation

When the installed CLI is accessible, start with:

Get-Command multisim-mcp
multisim-mcp --json doctor

Read full_workflow_ready and each stable checks[].id. Apply the provided repair instructions instead of repeatedly attempting COM activation. Use doctor --json --strict only in CI or when a non-zero incomplete-setup status is useful. Run doctor --json --connect only when the user wants a real COM and license probe; it may start Multisim.

Preview client configuration when requested:

multisim-mcp config --client claude-desktop --python C:\Python32\python.exe
multisim-mcp config --client codex --python C:\Python32\python.exe
multisim-mcp config --client generic --python C:\Python32\python.exe

Do not write over a live client configuration. Generate a fragment, inspect it, and merge it manually. Keep the MCP server local over stdio.

After the client connects, call runtime_status before the first experiment on a new installation. Prefer the high-level run_circuit_experiment tool whenever the requested schematic uses its supported component subset; it keeps the generated design, Multisim simulation, exported data, plot, and report tied to one source netlist.

For a new design, use the planning gate before writing a netlist: call plan_design_options, wait for the user's plan_id/option_id choice, then call select_design_option, prepare_design_specification, and (after the specification approval) prepare_netlist_draft. Resolve candidate families with resolve_component_requirements; once concrete ratings and model provenance have been reviewed, call approve_component_resolution. That approval only authorizes a later compiler and never by itself creates SPICE, files, a schematic, or a simulation.

For the bounded signal-passive template, call compile_executable_netlist next and show its pin-level CircuitDesign, calculated values, and SPICE preview. Before any schematic or experiment stage, call approve_executable_netlist with explicit component/topology/value/SPICE confirmations. This approval only authorizes schematic planning; it does not authorize file writes, stimuli, analysis commands, or simulation. To generate the approved schematic, pass the complete compiler response as executable_netlist, the approval artifact as netlist_approval, and the preview's exact spice_netlist as netlist to create_schematic_from_netlist; it revalidates the handoff and rejects changed SPICE text. Then call approve_simulation_plan with the same preview and netlist approval plus the reviewed ExperimentSpec (safe commands, measurements, and limits). Pass all three artifacts to run_verified_circuit_experiment; it revalidates the netlist, commands, and measurement contract before creating the schematic or starting Multisim. When the workbench handoff JSON is available, multisim-mcp execute-handoff --handoff --root --json provides a validation-only path; after explicit user confirmation, add --confirm to execute schematic generation followed by the verified experiment. It rejects root escapes, mismatched approvals, and existing artifacts by default. For a long-running handoff, use --submit --confirm after validation; the CLI creates the approved schematic first and then queues the verified experiment for the durable worker. For a durable long-running job, pass the same three artifacts and reviewed requirements to submit_circuit_experiment; the isolated worker persists and revalidates them before execution. After an approved run, verify the resulting directory.manifest.json through inspect-project. The manifest should contain only the sanitized approval_provenance identity, and a workbench result refresh should match its path, integrity, and approval/netlist/ compiled/spec digests before treating it as the current run. Legacy or direct experiments without that identity remain evidence-only.

1. Clarify the design

Collect from the user:

• Circuit function and constraints (input, output, supply, frequency).
• Component set and tolerance requirements.
• Required analyses: DC operating point, AC sweep, transient, or all.
• Report format and whether schematic image, netlist, BOM, or CSV data are needed.

2. Build and validate the netlist

• Write a SPICE netlist as the source of truth.
• Pre-validate with ngspice when available before touching Multisim.
• Keep node names simple and unique: vin, vout, vdd, vb, 0 work, while reserved words such as in/out can fail in the command engine.

3. Create or edit the design

Preferred paths, in order:

1. Start from a known-good template .ms14 and edit its decoded XML.
2. Use decode_ms14 to get XML, edit component values carefully (RLC values require updating both the numeric parameter and display string).
3. Use encode_ms14 to re-encode.
4. Use open_circuit to load the result.

Do not try to rename RefDes by editing display text; Multisim renumbers on open.

Netlist-driven alternative

When building a circuit from scratch, run_spice_netlist is the fastest path:

• Pass a SPICE netlist plus Nutmeg commands such as dc VIN 0 10 0.1 or tran 1u 2m.
• Multisim's command engine needs a space-free directory; the tool creates one automatically under MULTISIM_MCP_WORKDIR (default C:\msre_exp).
• The tool appends write without a variable list; write path v(out) can fail with No such vector.
• It waits for the engine to go idle, parses the SPICE3 raw file, and returns columns, n_points, sampled rows, and a CSV path.
• Use output_dir to copy circuit.cir, run.log, result.raw, and data.csv into the workspace.
• Only use the safe op, dc, ac, or tran command subset. Never request unrestricted command mode unless the user explicitly asks for trusted local security research and the server operator enabled it.

High-level generated experiment

For supported RLC, independent sources, diode, BJT, MOSFET, and op-amp circuits:

1. Call run_circuit_experiment with the validated netlist, analysis command, output directory, and report title.
2. Treat the returned .ms14, schematic PNG, raw, CSV, SVG, and Markdown as one experiment bundle.
3. Verify success, n_points, and expected numeric relationships before reporting completion.
4. Generated schematic probes are experimental; use command-engine CSV/raw data as the authoritative experiment result.

4. Run simulations

• Call connect, then open_circuit.
• connect is idempotent; call it at the start of the workflow even if another step already connected.
• Call enum_outputs and enum_inputs to discover valid names.
• Use run_dc_operating_point, run_ac_sweep, run_ac_single_frequency, or run_transient.
• Use run_spice_netlist when the design is netlist-only or a schematic does not exist yet; call connect (and new_circuit if needed) first.
• If a waveform input is required, use set_input_data_sampled or set_input_data_raw before the transient.
• If a tool reports not ready, call stop_simulation, check circuit_info, and retry once.

5. Analyze data

Returned data shape depends on the analysis:

• DC: two rows (value,).
• AC: three rows (frequency, real, imag).
• Transient: two rows (time, real).

Compute mean, min, max, rise/fall, bandwidth, gain, or FFT as required. Keep the raw data in a file when the user asks for a report. run_spice_netlist returns its own parsed columns and CSV; use those directly when the schematic-level analysis tools are not applicable.

When a concrete reversible DesignPatch has already been proposed, prefer evaluate_design_patch over an informal rerun. It evaluates the unchanged baseline and exactly one in-memory candidate under the same hard requirements, retains before/after diagnoses and the inverse patch, and never applies the candidate automatically. Treat adoption_eligible as evidence for a separate approval step, not as permission to overwrite the design.

6. Produce the report

• Export get_circuit_image, report_netlist, and report_bom.
• Use generate_report with the analysis dicts returned by run_dc_operating_point, run_ac_sweep, and run_transient to write a Markdown report with circuit info, exports, and measured tables.
• For run_spice_netlist, write the report from the returned CSV/rows and include the netlist, analysis commands, measured tables, and plots.
• Write a Markdown or CSV report with circuit description, component table, simulation setup, measured values, plots, and conclusions.
• Cite measured data and flag anything that was estimated or not verified.
• Enumeration tools return structured dicts such as {"outputs": [...]}; read the list from the outputs/inputs/components key.

Safety

• Always stop simulations before changing inputs or saving.
• Prefer a working directory for outputs; never overwrite the user's source design without save_circuit to a new path.
• Do not set unsafe_commands or invoke do_command_line in ordinary circuit workflows. Those capabilities are disabled by default for prompt-injection resistance.
• This project is unofficial and not affiliated with NI.