--- name: mps-language-analysis description: Analyze an MPS language by name — discover concepts, properties, references, children, aspects (editor/constraints/behavior), metadata, and answer type questions (is A a subtype of B, common supertype, expected type of a slot, why a node has its type) with `mps_mcp_query_types`. Use when investigating an unfamiliar language, exploring concept structure, or finding sample nodes to use as templates for JSON blueprints. type: reference --- # MPS Language Analysis ## Loading companion skills Companion names in this skill are lazy dependencies: load only those relevant to the current task. If this skill came from an MCP server, use the host's skill loader to resolve the companion's unique discovered entry URI on the same host-assigned originating server. If the host has no server-backed skill loader, stop and report that limitation; do not silently fall back to a filesystem copy. If this skill came from a filesystem catalog, load the named sibling from that same catalog at `//SKILL.md`, even if remote skill loaders are also available. Do not invent a tool name or server endpoint. Workflow for inspecting an MPS language from a name (e.g. `jetbrains.mps.lang.core`). Returns concepts, metadata, structural info, and pointers to declarations and sample nodes. ## Critical Directives - Use the **fully qualified** language name (e.g. `jetbrains.mps.lang.core`) — single-letter shorthand (`j.m.l.core`) requires resolution first via `mps_mcp_get_project_structure`. - For the `qualifiedName` returned by `mps_mcp_get_concept_details`, use it as the `concept` field in JSON blueprints. It is unambiguous. ## Analyzing a Language by Name 1. **Verify Language**: call `mps_mcp_get_project_structure` with the language name as a filter (`startingPoint: My_Language`). This confirms existence and provides the UUID. 2. **Retrieve Concepts**: use `mps_mcp_get_concept_details` with the language name in `languageRefs`. Both `conceptRefs` and `languageRefs` accept either one value or a JSON array (real or written as a string); omit the unused selector. 3. **Extract Data**: the response includes: * **Name**: concept FQN. * **Description**: found in `shortDescription`. * **Metadata**: `isRootable`, `isAbstract`, and the `conceptReference` ID. * **Structure**: properties, children, and references are detailed here. 4. **Drill Down**: * **Declaration**: use the `sourceNode` reference with `mps_mcp_open_node` to open the definition. * **Examples**: use `mps_mcp_query_nodes` with `FIND_INSTANCES` (`sampleOnly: true`) to get a sample node. Then use `mps_mcp_print_node` to see its canonical JSON structure for use as a template. * **Inheritance**: load the `mps-language-inheritance` skill for deeper hierarchy analysis. ## Inspecting Concept Aspects Use `mps_mcp_query_structure` with `LIST_CONCEPT_ASPECTS` to find associated definitions (editor, constraints, behavior, typesystem rules, generator templates): * **`relation`**: `appliesTo` marks a root that is *for* the concept (its editor, a rule applicable to it); `mentions` a root that only references it. `pattern: true` marks a rule that matches the concept through a pattern, so only instances of that shape. Details: `references/structure-operation-api/list-concept-aspects.md` in the `mps-aspect-structure-concepts` skill root after loading that companion skill from the same origin. * **Inherited Aspects**: set `includeInherited: true` to add the aspects of ancestors (superconcepts/interfaces); `targetsConcept` says which concept each root was found through. * **Editor Analysis**: * **ConceptEditorDeclaration**: with `relation: appliesTo`, the full editor for the `targetsConcept`. * **EditorComponentDeclaration**: defines reusable presentation pieces. * Check the `editor` model in the response to identify available editors. ## Answering Type Questions Let the typesystem engine answer instead of reading rule roots: a node's type → `mps_mcp_query_nodes` `GET_TYPE`; the type its context requires → `GET_EXPECTED_TYPE`; does A fit B → `mps_mcp_query_types` `IS_SUBTYPE`; what A widens to → `SUPERTYPES`; what A and B share → `COMMON_SUPERTYPE`; A seen as a given type concept → `COERCE`; the type of `a op b` → `OPERATION_TYPE`; why a node has its type → `EXPLAIN_TYPE`. Rule roots cannot tell you what replacement, comparison or overloaded-operation rules decide, and a rule's name need not match what it returns. `LIST_CONCEPT_ASPECTS` lists typesystem rules for authoring them; do not open a `SubtypingRule` or `InferenceRule` root to answer a type question. Operand forms and outputs: `references/analysis-tools/type-queries.md` in the `mps-mcp-workflow` skill root after loading that companion skill from the same origin. ## Related Skills - **`mps-language-inheritance`** — load when you need extended-language / superconcept / subconcept analysis. - **`mps-aspect-structure-concepts`** — load when defining or modifying concepts (not just reading them). - **`mps-language-aspects-overview`** — overview of which aspects exist and what each owns. ## Reference Index - Open `references/search-concepts.md` for the `mps_mcp_search_concepts` matching algorithm — haystack composition, subtoken splitting, fallback ranking, the three scopes (`modelReference` / `scope`) and the auto-widen, the `detail` projections, the 50-match cap, and the sub-2-char failure mode. - Open `references/concept-details.md` for the `mps_mcp_get_concept_details` result schema and the unresolved-ref policy (all-failed vs partial-success envelopes and the suggestion heuristic). ## Scripts `scripts/concept_shape.py` — reduces a `mps_mcp_get_concept_details` result file to one line per property, reference, and child role (type, enum literals, cardinality): the concept shape needed to author a blueprint, instead of the 10–40 KB details file. ``` python3 scripts/concept_shape.py /var/folders/.../mps-node-456.json --concept Course Course com.example.courses.structure.Course rootable prop credits integer prop level enum Level [INTRO|CORE|ADVANCED] child lessons com.example.courses.structure.Lesson 1..n child prerequisites com.example.courses.structure.CourseRef 0..n ``` `--all` keeps the inherited `shortDescription` / `virtualPackage` / `smodelAttribute` features that are hidden by default; `--list-tools` prints the tools and parameters it depends on. The script is a thin front end over `scripts/mps_dump.py` in the `mps-mcp-workflow` skill root after loading that companion skill from the same origin, which must be installed alongside this one; that library also offers the `roots`, `node`, and `count` projections. No `python3` (typically Windows): read the details file with the file reader and keep, per concept, only `qualifiedName`, `isAbstract`/`isRootable`, and for each entry of `properties` / `references` / `children` the `name`, `type`/`targetConcept`, `cardinality`, and `enumerationValues` (plus `enumerationDefault`, the literal a property holding the default value carries) — ignore `featureId`, `sourceNode`, `doc`, and `sampleNode` unless you need them.