MCP Server & IDE Workbench Reference

ModelScript provides first-class AI agent integration via the Model Context Protocol (@modelscript/mcp) alongside a comprehensive polyglot VS Code Extension & Web IDE (@modelscript/ide-client, @modelscript/lsp). Both surfaces share the same zero-allocation WebAssembly GLR parsers, Salsa QueryEngine, and linear-memory DAEBuilder.

Part I: Model Context Protocol (MCP) Server

The @modelscript/mcp package exposes ModelScript's compilation, simulation, formal verification, ontology reasoning, Triple Graph Grammar (TGG) synchronization, and requirements traceability engines to MCP-compatible AI coding assistants (Claude Desktop, Cursor, Windsurf, Antigravity, Zed, and custom LLM orchestrators).

1. Deployment Modes & Transports

ModelScript supports three MCP execution architectures:

A. Local `stdio` Server (`@modelscript/mcp`)

For local desktop AI agents, run @modelscript/mcp over standard input/output (StdioServerTransport):

VS Code (.vscode/mcp.json):

json

{
  "servers": {
    "modelscript": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelscript/mcp"],
      "env": {
        "MODELICAPATH": "/path/to/ModelicaStandardLibrary"
      }
    }
  }
}

VS Code CLI & Claude Code CLI:

bash

# VS Code (adds to user/workspace MCP configuration)
code --add-mcp '{"name":"modelscript","command":"npx","args":["-y","@modelscript/mcp"]}'

# Claude Code
claude mcp add modelscript -- npx -y @modelscript/mcp

Cursor (.cursor/mcp.json), Windsurf & Claude Desktop (claude_desktop_config.json):

json

{
  "mcpServers": {
    "modelscript": {
      "command": "npx",
      "args": ["-y", "@modelscript/mcp"],
      "env": {
        "MODELICAPATH": "/path/to/ModelicaStandardLibrary"
      }
    }
  }
}

B. Hosted Enterprise SSE Gateway (`/api/v1/mcp`)

The ModelScript Enterprise Hub (apps/api/src/routes/mcp.ts) exposes a multi-tenant HTTP Server-Sent Events (SSE) MCP gateway with built-in compute credit metering:

GET /api/v1/mcp/sse: Establishes a persistent SSE stream (text/event-stream), authenticates the caller via Authorization: Bearer <JWT | ms_bot_*>, and returns a session-scoped endpoint event (/api/v1/mcp/messages?sessionId=<uuid>).
POST /api/v1/mcp/messages?sessionId=<uuid>: Receives JSON-RPC 2.0 tools/call and resources/read requests, checks organization compute credit balances via BillingService (1 credit per lightweight query, 5 credits for modelica_simulate / hybrid_simulate / modelica_doe / modelscript_closed_loop_synthesize), and streams results over the SSE channel.

C. In-WASM AOT Polyglot Dispatch (`PolyglotMcpHost`)

In browser workers and edge sandboxes (packages/mcp/src/polyglot-server.ts), PolyglotMcpHost dynamically synthesizes MCP tool schemas from every registered LanguageDefinition (defineLanguage({ mcp: { tools, resources } })). When a tool declares an aotExport symbol, PolyglotMcpHost writes the arguments directly into WebAssembly linear memory (__newString), invokes the compiled AssemblyScript kernel (mcpDispatchTool(toolId, targetPtr, paramsJsonPtr)), and reads back the JSON result with zero JS object marshaling overhead.

2. Complete Reference of All 52 Core MCP Tools

The canonical MCP server (packages/mcp/src/tools.ts) registers 52 specialized engineering tools grouped across nine functional domains:

Domain 1: Modelica Compilation, Simulation, Linearization & Sensitivity (11 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`modelica_load`	`path: string`	—	Load a `.mo` file or package directory from disk into the stateful MCP workspace session and return all discovered class FQNs.
`modelica_parse`	—	`code?: string`, `path?: string`	Parse Modelica source text or a file using the native WASM GLR parser and return CST syntax diagnostics and top-level class names.
`modelica_flatten`	`className?: string`	`name?: string`, `code?: string`, `path?: string`, `includeMetrics?: boolean`	Flatten a hierarchical Modelica class into a canonical DAE (`ArenaDAEPrinter` output) and return structural BLT/Pantelides metrics and flattened source text.
`modelica_lint`	—	`code?: string`, `path?: string`	Execute all `M1xxx`–`M6xxx` static lint rules and Salsa semantic queries, returning structured diagnostics with line/column spans.
`modelica_simulate`	`className: string`	`code?: string`, `path?: string`, `tStart?: number` (`0`), `tEnd?: number` (`1`), `dt?: number` (`0.01`), `tolerance?: number` (`1e-6`), `method?: "euler" \| "rk4" \| "dopri5" \| "bdf" \| "radau5" \| "cvode" \| "ida"` (`"rk4"`), `variables?: string[]`	Flatten, index-reduce (Pantelides + BLT tearing), and numerically simulate a Modelica model, returning downsampled time-series trajectories and solver statistics.
`hybrid_simulate`	`className: string`	`code?: string`, `path?: string`, `tStart?: number` (`0`), `tEnd?: number` (`10`), `dt?: number` (`0.01`), `maxEvents?: number` (`500`)	Run hybrid continuous/discrete event-driven simulation with zero-crossing root finding (`when` clauses, `reinit()`, state machines, and synchronous clocks).
`modelica_doe`	`className: string`, `parameters: Record<string, number[]>`	`code?: string`, `path?: string`, `tEnd?: number` (`1`), `dt?: number` (`0.01`), `metrics?: string[]`	Execute a Cartesian/factorial Design of Experiments (DoE) sweep across parameter combinations and return terminal/summary metrics for each run.
`modelica_sensitivity`	`className: string`, `parameters: string[]`, `outputs: string[]`	`code?: string`, `path?: string`, `tEnd?: number` (`1`), `dt?: number` (`0.01`), `perturbation?: number` (`0.01`)	Compute normalized local sensitivity Jacobians ( $\frac{\partial y_{i}}{\partial p_{j}} \cdot \frac{p_{j}}{y_{i}}$ ) via central finite differences or forward AD.
`modelica_query`	`queryType: "list_classes" \| "class_info" \| "components" \| "equations" \| "connections"`	`className?: string`	Query the active Salsa `SymbolIndex` for class specializations, typed component declarations, equations, or `connect()` topology edges.
`modelica_linearize`	`className: string`	`code?: string`, `operatingPoint?: Record<string, number>`, `inputs?: string[]`, `outputs?: string[]`	Linearize a Modelica DAE around an equilibrium point to extract explicit LTI State-Space matrices ( $A, B, C, D$ ), eigenvalues, damping ratios, and stability margins.
`modelica_tune_pid`	`className: string`	`code?: string`, `controllerType?: "P" \| "PI" \| "PID"`, `targetBandwidthRadPerSec?: number`, `maxSensitivityMs?: number`	Synthesize $K_{p}, K_{i}, K_{d}, N$ controller gains for a linearized Modelica plant using Skogestad IMC (SIMC) and sensitivity-constrained loop shaping.

Domain 2: Cross-Domain Consistency & Ontology Reasoning (5 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`validate_system_consistency`	—	`sysmlCode?: string`, `modelicaCode?: string`, `sysmlPath?: string`, `modelicaPath?: string`, `mapping?: Record<string, string>`	Cross-check a SysML v2 architecture definition against a Modelica physics implementation for port mismatches, unit conflicts, and shared parameter divergences.
`query_ontology_sparql`	`ontologyText: string`, `sparqlQuery: string`	`enableReasoning?: boolean` (`true`)	Parse an OWL 2 Functional Syntax or Turtle ontology, optionally run OWL 2 RL forward-chaining inference, and execute a SPARQL 1.1 `SELECT`/`ASK` query.
`query_ontology_bgp`	`ontologyText: string`, `patterns: Array<{subject?: string, predicate?: string, object?: string}>`	`enableReasoning?: boolean` (`true`)	Match Basic Graph Patterns (BGP) against the inferred RDF triple store and return variable bindings.
`trace_fault_propagation`	`ontologyText: string`, `faultSourceIri: string`	`propagationPropertyIri?: string`, `maxDepth?: number` (`10`)	Traverse transitive causal/containment relationships across the ontological knowledge graph to compute downstream failure blast radius and FMEA impact chains.
`explain_inference`	`ontologyText: string`, `subject: string`, `predicate: string`, `object: string`	—	Construct a formal proof derivation tree explaining which OWL 2 RL rules and asserted axioms entailed a specific inferred triple $(s, p, o)$ .

Domain 3: Triple Graph Grammar (TGG) Digital Thread Synchronization (3 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`tgg_query_thread`	`sysmlCode: string`, `modelicaCode: string`	`spiceCode?: string`	Build the multi-domain Triple Graph Grammar correspondence graph across SysML v2, Modelica, and SPICE, returning synchronized element triples and orphan nodes.
`tgg_diagnose_conflict`	`sysmlCode: string`, `modelicaCode: string`	`baseSysmlCode?: string`, `baseModelicaCode?: string`	Run Critical Pair Analysis (CPA) over concurrent modifications in SysML v2 and Modelica to detect conflicting edits on shared parameters or ports.
`tgg_reconcile_slot`	—	`conflictId?: string`, `slotId?: string`, `correspondenceId?: string`, `strategy?: string`, `authority?: "sysml" \| "modelica"`, `customValue?: number`, `sysmlCode?: string`, `modelicaCode?: string`, `sysmlPath?: string`, `modelicaPath?: string`	Resolve a parameter or structural divergence on a TGG slot by propagating the authoritative or physics-constrained value across SysML v2 and Modelica.

Domain 4: Requirements Traceability Matrix (RTM) Intelligence (3 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`rtm_query_matrix`	`files: Record<string, string>`	—	Extract all requirements, architectural elements, physics models, and verification cases across a polyglot file map and compute coverage percentages and orphan counts.
`rtm_suggest_links`	`files: Record<string, string>`	`threshold?: number` (`0.35`)	Analyze identifier lexical similarity, shared physical quantities/units, and structural proximity to recommend missing `@satisfies` and `@verifies` traceability links.
`rtm_diagnose_suspect`	`baseFiles: Record<string, string>`, `headFiles: Record<string, string>`	—	Compare two workspace snapshots via semantic AST fingerprinting to flag downstream `@satisfies`/`@verifies` links as `SUSPECT` when upstream requirement expressions change.

Domain 5: Autonomous Closed-Loop Synthesis, Calibration & Workspace Editing (5 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`modelscript_closed_loop_synthesize`	`code: string`, `language: "sysml2" \| "modelica"`	`targetModelName?: string`, `filePath?: string`, `simulateAndTune?: boolean`, `tEnd?: number`, `tunableParameters?: string[]`, `requirements?: Array<{variable: string, condition: "lt" \| "gt" \| "eq" \| "between", threshold: number, upperThreshold?: number}>`	Execute an autonomous 5-gate compile $\to$ QUDV $\to$ SMT $\to$ DAE balance $\to$ dynamic simulation, STL robustness ( $ρ$ ), and parameter tuning loop.
`modelscript_patch_code`	`filePath: string`, `search: string`, `replace: string`	`validateBeforeWrite?: boolean` (`true`)	Apply a surgical search-and-replace edit to a workspace file, automatically running the WASM GLR parser and linter before committing the write to disk.
`modelscript_read_file`	`filePath: string`	`startLine?: number`, `endLine?: number`	Read source lines from a workspace file within the sandboxed root directory.
`modelscript_list_workspace_files`	—	`directory?: string`, `extensions?: string[]`	Recursively list engineering source files (`.mo`, `.sysml`, `.msx`, `.pscx`, `.scad`, `.stp`, `.owl`, `.ssp`, `.csv`) in the active workspace.
`modelica_diff_calibrate`	`className: string`, `modelicaCode: string`, `parameters: string[]`, `referenceData: {time: number[], signals: Record<string, number[]>}`	`learningRate?: number` (`0.05`), `maxEpochs?: number` (`50`), `parameterBounds?: Record<string, [number, number]>`	Calibrate unknown physical parameters against experimental CSV/reference trajectories using differentiable simulation and gradient descent (Adam / L-BFGS) to minimize trajectory MSE.

Domain 6: PSCAD Power Systems & IEEE 2800 Grid Compliance (4 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`pscad_lint`	—	`path?: string`, `fileName?: string`, `code?: string`, `xmlContent?: string`	Run all 11 electrical EMT lint rules (`M6001`–`M6011`) on a `.pscx`, `.pslx`, or `.pswx` XML payload.
`pscad_simulate`	`xmlContent: string`	`tStop?: number` (`1.0`), `stepUs?: number` (`20.0`)	Execute a fixed-step nodal admittance EMT simulation on a `.pscx` case and return three-phase bus voltage, branch current, and CSMF control signal waveforms.
`pscad_verify_ieee2800`	—	`path?: string`, `fileName?: string`, `code?: string`, `xmlContent?: string`, `tStop?: number`, `stepUs?: number`, `faultBus?: string`, `retainedVoltagePu?: number`	Run automated IEEE Std 2800-2022 LVRT/HVRT voltage ride-through, frequency ride-through, and reactive current injection compliance verification.
`pscad_export`	`xmlContent: string`, `targetFormat: "modelica" \| "ssp" \| "svg"`	—	Convert a PSCAD `.pscx` schematic into an executable Modelica `.mo` model, an SSP 1.0 `.ssd` descriptor, or a rendered Single-Line Diagram SVG.

Tool Name	Required Arguments	Optional Arguments	Description
`social_read_thread`	`postId: number`	—	Read a social discussion thread with its full conversation tree, parent posts, child replies, and attached engineering artifact views.
`social_post_reply`	`replyToId: number`, `content: string`	`artifactViewId?: number`	Post an engineering response or simulation synthesis back into a discussion thread on behalf of an authorized bot or automated agent.

Domain 8: Multibody Rail Dynamics & Multiphase Flow Assurance (14 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`olga_simulate`	—	`path?: string`, `code?: string`, `tEnd?: number`, `dt?: number`	Simulate transient multiphase flow, liquid accumulation, pressure drops, slugging, and pigging for an OLGA model (`.opi`, `.key`, `.genkey`).
`olga_lint`	—	`path?: string`, `code?: string`	Run pre-flight multiphase flow assurance diagnostics (`M7001`–`M7007`: CFL stability, steep risers, severe slugging dips, pig stuck hazard) on an OLGA model.
`olga_verify`	—	`path?: string`, `code?: string`, `maxPressureBar?: number`, `minHydrateMarginC?: number`, `maxSlugVolumeM3?: number`	Formally verify multiphase simulation against flow assurance limits (MAOP, hydrate subcooling margin, and slug catcher capacity).
`olga_diagram`	—	`path?: string`, `code?: string`, `view?: "pid" \| "profile"`	Generate standalone SVG diagrams for an OLGA flowline: topological P&ID schematic or pipeline elevation bathymetry $Z (X)$ with holdup heatmaps.
`olga_parse_tpl`	—	`path?: string`, `content?: string`	Parse an OLGA simulation trend plot (`.tpl`) file or payload into a structured time-series dataset with variable catalog, units, and channel statistics.
`olga_parse_ppl`	—	`path?: string`, `content?: string`	Parse an OLGA spatial profile plot (`.ppl`) file or payload into structured spatial snapshots along pipeline coordinate $X$ with flow regimes and extrema.
`olga_tune`	—	`path?: string`, `code?: string`, `telemetryCsv?: string`, `parameters?: string[]`	Calibrate OLGA flowline model parameters (roughness, choke $C_{d}$ , heat transfer, well PI) using Nelder-Mead simplex history matching against telemetry observations.
`olga_gate`	—	`path?: string`, `code?: string`	Evaluate an automated Flow Assurance CI/CD Safety Gate check on an OLGA model for pull requests and operational envelope verification.
`olga_survey`	`csvText: string`	`maxSegments?: number`, `roughnessMm?: number`, `diameterM?: number`	Ingest 3D seabed bathymetry survey coordinates (CSV with Easting/Northing/Depth or KP/Elevation) and discretize into curvature-aware OLGA pipe segments.
`gensys_lint`	—	`path?: string`, `code?: string`	Perform pre-flight semantic and physical validation on GENSYS railway vehicle models (`.calc`, `.run`, `.sub`).
`gensys_verify_en14363`	—	`path?: string`, `code?: string`, `speedKmh?: number`, `cantDeficiencyMm?: number`	Verify Nadal derailment quotient ( $Y / Q \leq 0.8$ ) and Prud'homme lateral track shifting force limits according to EN 14363 / UIC 518.
`gensys_fastsim`	`normalForceN: number`, `semiAxisAMm: number`, `semiAxisBMm: number`	`creepX?: number`, `creepY?: number`, `spinRadPerM?: number`, `frictionCoeff?: number`	Compute tangential longitudinal/lateral creep forces and spin moment using Kalker's FASTSIM wheel-rail contact algorithm.
`simpack_lint`	—	`path?: string`, `code?: string`	Perform structural and reference diagnostics on SIMPACK XML (`.spck`) or classic deck (`.sim`, `.mbs`) multibody models.
`simpack_mobility`	—	`path?: string`, `code?: string`	Calculate spatial system degrees of freedom and detect overconstrained mechanisms using the Grübler-Kutzbach mobility criterion.

Domain 9: Cross-Domain CAD/FEA Sizing, B2B Migration, Co-Sim Stability, Formal Safety & Digital Twin (5 Tools)

Tool Name	Required Arguments	Optional Arguments	Description
`cad_fea_closed_loop_size`	—	`scadCode?: string`, `stepPath?: string`, `material?: string`, `yieldStrengthMPa?: number`, `minSafetyFactor?: number`, `appliedLoadN?: number`, `tunableDimensions?: Record<string, number>`, `clearanceEnvelopeMm?: number`	Run closed-loop 3D CAD/OpenSCAD/STEP parametric sizing coupled with tetrahedral FEA von Mises stress analysis and 3D BVH collision clearance verification.
`polyglot_migrate_and_verify_b2b`	`sourceFormat: "simulink" \| "spice" \| "adams"`	`sourceCode?: string`, `sourcePath?: string`, `targetModelName?: string`, `generateSysmlWrapper?: boolean`, `verifyC99Codegen?: boolean`, `toleranceRmse?: number`, `tEnd?: number`	Migrate legacy Simulink (`.slx`), SPICE (`.cir`), or MSC Adams (`.adm`) models to Modelica & SysML v2 and certify ISO 26262 Back-to-Back (B2B) trajectory & C99 parity.
`ssp_diagnose_cosim_stability`	—	`ssdXml?: string`, `sspPath?: string`, `macroStepSizeSec?: number`, `couplingMethod?: "jacobi" \| "gauss-seidel" \| "tlm" \| "ijcsa"`	Analyze an FMI 2.0/3.0 & SSP multi-FMU co-simulation topology for algebraic feedthrough loops and ZOH spectral radius ( $ρ (J_{ZOH})$ ) instability.
`formal_verify_safety_and_mcdc`	`code: string`	`className?: string`, `safetyProperty?: string`, `engine?: "ic3-pdr" \| "k-induction" \| "sos-barrier"`, `maxDepth?: number`, `generateMcdcVectors?: boolean`	Prove unbounded safety invariants via IC3/PDR and Sum-of-Squares (SOS) barrier certificates, and generate DO-178C / ISO 26262 MC/DC independence-pair test vectors.
`twin_discover_sindy_and_rom`	—	`className?: string`, `modelicaCode?: string`, `telemetryData?: {time: number[], states: Record<string, number[]>}`, `sparsityThreshold?: number`, `romRank?: number`	Discover sparse nonlinear governing differential equations (SINDy STLSQ) from hardware telemetry residuals and calibrate a Reduced-Order Model (ROM) digital twin.

Declarative Polyglot Language Tools (`PolyglotMcpHost`)

When using PolyglotMcpHost (packages/mcp/src/polyglot-server.ts), additional language-declared tools from languages/*/src/language.ts are automatically mounted:

sysml2_extract_topology: Extracts hierarchical part definitions, ports, and item flow connections from a .sysml workspace symbol (aotExport: "mcp_sysml2_extract_topology").
sysml2_verify_requirements: Evaluates SysML v2 requirement def constraint satisfaction across the workspace (aotExport: "mcp_sysml2_verify_requirements").

3. MCP Resources (`resources/list` & `resources/read`)

The MCP server exposes live workspace state as URI-addressable resources (packages/mcp/src/resources.ts and languages/*/src/language.ts):

Resource URI / Template	MIME Type	Description
`modelica://libraries`	`application/json`	Lists all loaded Modelica package roots and top-level libraries in the current session.
`modelica://classes`	`application/json`	Lists every indexed class FQN, specialized kind (`model`, `block`, `connector`, `record`, `function`, `package`), and description.
`modelica://classes/{name}`	`application/json`	Returns full class metadata for `{name}`, including source text, components, equations, and `connect()` clauses.
`modelica://workspace/{className}/ast`	`application/json`	(`PolyglotMcpHost`) Serializes the linear CST/AST structure of `{className}`.
`modelica://workspace/{className}/dae`	`application/json`	(`PolyglotMcpHost`) Returns the lowered DAE variables, parameters, and equations for `{className}`.
`sysml2://workspace/{partName}/topology`	`application/json`	(`PolyglotMcpHost`) Returns the resolved port and connection graph for a SysML v2 part.

Part II: VS Code Extension & Web IDE Workbench

The official ModelScript extension (modelscript.modelscript, source in apps/ide/ and packages/lsp/) runs both in VS Code Desktop (Node.js extension host) and VS Code for the Web / ide.modelscript.org (100% browser WebWorker host).

1. Polyglot Language Server (`@modelscript/lsp`)

A single unified Language Server Protocol instance provides incremental language intelligence across the registered languages (.mo, .sysml, .kerml, .msx, .pscx, .pswx, .scad, .stp, .step, .owl, .ssp, .ssd, .ssv, .csv, .su2, .cfg, .inp, .gbr):

Real-Time Diagnostics: Incremental GLR parsing + Salsa lint__* queries + cross-file unit/type checking.
Navigation & Refactoring: Go-to-Definition, Find All References (across .sysml $\leftrightarrow$ .mo $\leftrightarrow$ .msx boundaries), Workspace Symbol Search, Rename Symbol, and Document Outline.
Code Lenses & Inlay Hints: Inline equation/variable balance badges ([Balanced: 14 vars / 14 eqs]), inferred SI unit inlay hints, and one-click ▶ Simulate, ⚡ Verify, and 📐 Render Diagram CodeLenses above every model, part def, and verification def.

2. Debug Adapter Protocol (DAP) Simulation Debugger

ModelScript registers a full Debug Adapter Protocol implementation ("type": "modelscript") that lets engineers step through continuous-time DAE integrations, state-machine transitions, and zero-crossing events exactly like stepping through imperative code:

Time & Condition Breakpoints: Set breakpoints on equation lines, when clauses, state transitions, or conditional simulation expressions (e.g., time >= 2.5 or der(omega) < 0).
Execution Controls: Continue, Step Over (advance one solver step $Δ t$ ), Step Into (step into Newton-Raphson / event-iteration micro-steps), Step Back (reverse-time trajectory scrubbing), and Restart.
Live Inspection Scopes: Inspect States (x), Derivatives (der(x)), Algebraic Variables (y), Parameters (p), Discrete / Clocked, and Solver Telemetry (h, localError, jacobianCond) in the VS Code Variables and Watch panes.

json

{
  "version": "0.2.0",
  "configurations": [
    {
      "type": "modelscript",
      "request": "launch",
      "name": "Debug Modelica Simulation",
      "model": "Modelica.Mechanics.Rotational.Examples.First",
      "program": "${file}",
      "stopTime": 5.0,
      "solver": "rodas4p",
      "breakOnZeroCrossing": true
    }
  ]
}

3. Virtual Filesystems & Interactive Notebooks (`.monb`)

Virtual FileSystem Providers:
- modelica://: Read-only browsable virtual filesystem mounting the bundled Modelica Standard Library (MSL) and installed registry packages (modelica_modules/).
- memfs://: Ephemeral in-memory workspace used in the browser IDE (ide.modelscript.org).
- github://: Zero-clone direct browsing and editing of remote GitHub repositories.
ModelScript Interactive Notebooks (modelscript.monbNotebook):
- Native VS Code NotebookSerializer and NotebookController for .monb files.
- Supports polyglot code cells (modelica, sysml2, modelscript, scad, sparql) that share a reactive kernel session, rendering interactive Bode plots, time-series charts, 3D CAD viewports, and SVG schematics directly below each cell.

4. Custom Document Editors & Engineering Workbench Panels

The extension contributes 9 custom visual editors and over 20 specialized engineering webview panels:

Custom Editor / View ID	Bound File Types / Trigger	Capabilities
`modelscript.diagramEditor`	`.mo`, `.sysml`, `.pscx`	Bi-directional graphical schematic & SysML v2 (BDD/IBD/State/Activity) editor; dragging components or routing wires updates the underlying source CST in real time.
`modelscript.monbNotebook`	`.monb`	Literate engineering notebook combining Markdown/LaTeX narratives with live simulation and verification cells.
`modelscript.fmuViewer`	`.fmu`	Inspects FMI 2.0/3.0 `modelDescription.xml`, causal ports, dependency sparsity matrices, and runs interactive test simulations on `.fmu` files.
`modelscript.sspEditor`	`.ssp`, `.ssd`, `.ssv`	Visual block-diagram editor for wiring multi-FMU co-simulation topologies and editing `.ssv` parameter tables.
`modelscript.reqifEditor`	`.reqif`, `.reqifz`	Tabular requirements specification editor with suspect-link indicators and one-click SysML v2 synchronization.
`modelscript.cadViewer`	`.stp`, `.step`, `.scad`, `.stl`, `.glb`	Hardware-accelerated Three.js / WebGL 3D CAD viewport with section planes, B-Rep face selection, and kinematic animation playback.
`modelscript.gerberViewer`	`.gbr`, `.drl`	Multi-layer 2D/3D PCB copper, solder mask, silkscreen, and drill stackup inspector.
`modelscript.cfdViewer`	`.su2`, `.vtk`, `.cfg`	Unstructured CFD mesh wireframe, boundary marker inspector, and pressure/velocity contour viewer.
`modelscript.feaViewer`	`.inp`, `.frd`	Finite-element mesh inspector with deformed shape magnification and von Mises stress colormaps.
Simulation & Bode Plot Studio	`modelscript.simulate`	Interactive multi-axis oscilloscope, FFT spectrum analyzer, phase portrait plotter, and linearized State-Space ( $A, B, C, D$ ) Bode/Nyquist root-locus analyzer.
BLT & Tearing Matrix Inspector	`modelscript.inspectBlt`	Visualizes the Tarjan Block Lower Triangular incidence matrix, Pantelides differentiated equations, and Cellier-Elmqvist tearing variables.
Formal Verification & Counterexample Center	`modelscript.verify`	Interactive dashboard for running BMC/ $k$ -induction/IC3/STL falsification and replaying falsifying counterexample traces step-by-step on the schematic.
Digital Thread & RTM Explorer	`modelscript.digitalThread`	Interactive graph explorer showing end-to-end traceability across OWL2 $\to$ SysML v2 $\to$ Modelica $\to$ CAD $\to$ Verification DHF.

5. Three-Tier AI Engineering Copilot

The IDE integrates an engineering-aware AI assistant (modelscript.aiChatView) that operates across three configurable inference tiers:

Tier 1 — Zero-Cloud Browser WebGPU (Qwen3-0.6B / WebLLM): Runs 100% locally inside the browser or desktop GPU using quantized WebGPU weights for air-gapped syntax completion, equation explanation, and unit conversion with zero data egress.
Tier 2 — Local Workstation Inference (Ollama / vLLM OpenAI-compatible): Connects to a local model server (http://localhost:11434) with automatic tool-calling bound to the in-process @modelscript/mcp tools.
Tier 3 — Hosted Frontier Models + MCP Gateway: Connects frontier reasoning models to the workspace's live @modelscript/mcp server (or /api/v1/mcp/sse), enabling autonomous closed-loop synthesis (modelscript_closed_loop_synthesize), differential calibration (modelica_diff_calibrate), and multi-domain TGG reconciliation (tgg_reconcile_slot).

ModelScript features native autonomous bot support across the social engineering network (apps/api and apps/web), allowing automated agents to participate in engineering discussions, verify physical equations, run continuous DAE simulations, and attach interactive artifact views.

1. Bot Registration & Scopes

Users can register autonomous bots in Settings $\to$ Autonomous Bots or programmatically via POST /api/v1/users/me/bots:

Dedicated Bot Identity: Bots receive a unique handle (e.g. @dynamics_bot), custom display name, bio, avatar, and system persona prompt.
ms_bot_* Bearer Tokens: Authenticate bot calls against social endpoints, the git repository engine, and the MCP SSE gateway.
Granular Capability Scopes:
- social:read: Read public engineering threads, parent contexts, and mentions.
- social:write: Post responses, explanations, and quotes into social threads.
- social:simulate: Execute DAE solvers (SUNDIALS CVODE/IDA) and synthesize interactive trajectory plots.
- modelica:validate: Compile and check Modelica CST syntax and static semantics.

2. Bring-Your-Own-Key (BYOK) Inference

Bots can run on instance-default models or configure private credentials:

BYOK Cloud API Keys: Supports Gemini, OpenAI, and Anthropic API keys directly configured on the bot.
Custom Inference Endpoints: Point to local Ollama (http://localhost:11434/v1), vLLM, or private enterprise LLM gateways.
Inference Models: Select from specialized models including gemini-1.5-flash, gemini-1.5-pro, gpt-4o-mini, gpt-4o, deepseek-r1, or qwen-2.5-coder.

3. Mention Processing & Live WebSocket Indicators

When an engineering post or comment mentions an automated bot (@dynamics_bot):

Background Queue: The API enqueues the job into bot_mention_queue with anti-loop recursion safeguards (bot mentions cannot trigger nested bot replies).
Real-Time WebSocket Indicators: The backend streams live status updates over /api/v1/collab/stream:
- thinking: The bot is parsing prompt equations and generating physical models.
- simulating: The bot is running numerical integrators and preparing trajectory data.
- done: The reply is posted and the UI automatically appends the completed card.
Interactive Artifact Synthesis:
- morsel: Inline interactive Modelica code view with syntax highlighting and verification badges.
- vega-plot: Interactive Vega-Lite timeseries charts with state variable selection and trajectory zoom.

MCP Server & IDE Workbench Reference ​

Part I: Model Context Protocol (MCP) Server ​

1. Deployment Modes & Transports ​

A. Local stdio Server (@modelscript/mcp) ​

B. Hosted Enterprise SSE Gateway (/api/v1/mcp) ​

C. In-WASM AOT Polyglot Dispatch (PolyglotMcpHost) ​

2. Complete Reference of All 52 Core MCP Tools ​

Domain 1: Modelica Compilation, Simulation, Linearization & Sensitivity (11 Tools) ​

Domain 2: Cross-Domain Consistency & Ontology Reasoning (5 Tools) ​

Domain 3: Triple Graph Grammar (TGG) Digital Thread Synchronization (3 Tools) ​

Domain 4: Requirements Traceability Matrix (RTM) Intelligence (3 Tools) ​

Domain 5: Autonomous Closed-Loop Synthesis, Calibration & Workspace Editing (5 Tools) ​

Domain 6: PSCAD Power Systems & IEEE 2800 Grid Compliance (4 Tools) ​

Domain 7: Social Engineering Network & Bot Collaboration (2 Tools) ​

Domain 8: Multibody Rail Dynamics & Multiphase Flow Assurance (14 Tools) ​

Domain 9: Cross-Domain CAD/FEA Sizing, B2B Migration, Co-Sim Stability, Formal Safety & Digital Twin (5 Tools) ​

Declarative Polyglot Language Tools (PolyglotMcpHost) ​

3. MCP Resources (resources/list & resources/read) ​

Part II: VS Code Extension & Web IDE Workbench ​

1. Polyglot Language Server (@modelscript/lsp) ​

2. Debug Adapter Protocol (DAP) Simulation Debugger ​

3. Virtual Filesystems & Interactive Notebooks (.monb) ​

4. Custom Document Editors & Engineering Workbench Panels ​

5. Three-Tier AI Engineering Copilot ​

Part III: Social AI Agents & Autonomous Bots ​

1. Bot Registration & Scopes ​

2. Bring-Your-Own-Key (BYOK) Inference ​

3. Mention Processing & Live WebSocket Indicators ​

MCP Server & IDE Workbench Reference

Part I: Model Context Protocol (MCP) Server

1. Deployment Modes & Transports

A. Local `stdio` Server (`@modelscript/mcp`)

B. Hosted Enterprise SSE Gateway (`/api/v1/mcp`)

C. In-WASM AOT Polyglot Dispatch (`PolyglotMcpHost`)

2. Complete Reference of All 52 Core MCP Tools

Domain 1: Modelica Compilation, Simulation, Linearization & Sensitivity (11 Tools)

Domain 2: Cross-Domain Consistency & Ontology Reasoning (5 Tools)

Domain 3: Triple Graph Grammar (TGG) Digital Thread Synchronization (3 Tools)

Domain 4: Requirements Traceability Matrix (RTM) Intelligence (3 Tools)

Domain 5: Autonomous Closed-Loop Synthesis, Calibration & Workspace Editing (5 Tools)

Domain 6: PSCAD Power Systems & IEEE 2800 Grid Compliance (4 Tools)

Domain 7: Social Engineering Network & Bot Collaboration (2 Tools)

Domain 8: Multibody Rail Dynamics & Multiphase Flow Assurance (14 Tools)

Domain 9: Cross-Domain CAD/FEA Sizing, B2B Migration, Co-Sim Stability, Formal Safety & Digital Twin (5 Tools)

Declarative Polyglot Language Tools (`PolyglotMcpHost`)

3. MCP Resources (`resources/list` & `resources/read`)

Part II: VS Code Extension & Web IDE Workbench

1. Polyglot Language Server (`@modelscript/lsp`)

2. Debug Adapter Protocol (DAP) Simulation Debugger

3. Virtual Filesystems & Interactive Notebooks (`.monb`)

4. Custom Document Editors & Engineering Workbench Panels

5. Three-Tier AI Engineering Copilot

Part III: Social AI Agents & Autonomous Bots

1. Bot Registration & Scopes

2. Bring-Your-Own-Key (BYOK) Inference

3. Mention Processing & Live WebSocket Indicators