Glossary

Terms from Sparx Enterprise Architect (EA), MDG Technologies, the AI clients and AI Power Tools, in alphabetical order. Westbrook Bank examples are illustrations: replace them with your own technology, stereotypes and model. Terms about the practice and NovoCircle's services are in the NovoCircle glossary.

A

Ad hoc stereotype. A stereotype applied to an element by name only, with no profile behind it. EA records the name but no profile. A name can be ad hoc on some elements and profile-bound on others, and the census reports that split. See ea-mdg-assess.

AI. One of the four labels that say who decides. The AI model decides, and results vary between runs and between models. See Where the AI decides and where AI Power Tools decides.

AI + AI Power Tools checks. The AI plans and drives, and deterministic or rule-based steps do the computing and checking. Most skills and every scenario work this way.

Alias (display term). The user-facing name a technology gives a stereotype or tag, such as "Business Application" for WBABusinessApplication. resolve_display_term looks one up. Tag aliases are usually not read from an MDG. See Display terms.

Application-level technology. A technology installed for the user or machine rather than stored in one model. Both an application-level and an embedded copy can exist for one id, and then the file wins. See Technology states.

Audit. The one implemented operation of ea_validate. It runs a ruleset against the model or one package and returns violations. See ea_validate.

B

Baseline. A snapshot of a package and everything beneath it, stored inside the repository. create_baseline takes one, compare_baseline returns a structured difference, and apply_baseline restores the package to it and discards later changes. A baseline is not a backup and is not version control. See Baselines and ea-change-management.

Binding (notation). A file of facts about one modeling language that the diagram engine uses: default sizes, spacing, routing, which layout grammar suits each diagram type, and the stereotype prefix. A diagram type with no binding is composed from engine defaults. See Notation bindings.

Binding (stereotype). The state of a stereotype that EA resolved to a profile, so the element takes the technology's shape and tagged values. A stereotype with no profile is ad hoc. summarize_stereotype_usage shows which is which.

Browser (EA). EA's project browser, the tree of packages and elements. A change written to the repository does not reach a Browser that already has the package cached, so it must be reloaded. Not the same as a web browser, which the Prolaborate skills use.

Browser control. The capability that lets the assistant operate web applications, through a built-in browser or Chrome. All Prolaborate skills depend on it. Computer use can see a browser but cannot click in one, so browser control is a separate capability. See Capabilities and surfaces.

Bulk operation. An operation whose name ends in _bulk. It applies a list of items, reports each outcome and does not stop at one failure. Creation bulks skip duplicates rather than updating them.

Business layer. The set of tables that the modeling technology governing a model defines over EA's own data: one entity table per element stereotype, one connector table per connector combination the technology allows, tag-row tables only where a column cannot hold the data, and a Table Directory. It sits over a scoped copy of EA's tables, or scoped views over them, and is the same on every delivery way: views in the EA database, a separate reporting database or Parquet files. The alignment check measures it against the technology's metamodel, and the value check compares every cell with an independent read through EA. A kept link into a left-out package ends at an "(excluded)" stand-in, which shows the element type and never the real name. See Reporting business layer and ea-reporting-database.

C

Census. A count of which stereotype each element and each connector carries and from which profile. It separates profile-bound, ad hoc and EA-internal stereotypes, compares what a technology declares with what the model uses, and feeds MDG assessment and the inclusion choice for reporting. It runs as a code engine.

Claude Code. Anthropic's coding tool, as the Code tab of the Claude desktop app and as a command-line tool in a terminal. The Code tab gets the server from the Claude Desktop extension and the skills from the plugin on your account. Claude Code in a terminal does not load Claude Desktop extensions. See Supported clients.

Claude Cowork. Claude's working mode that runs in the cloud by default. A cloud session has no local MCP server and cannot reach EA; a local session runs on your workstation with the extension's server beside your EA. Its skills come from the plugin on your account. Not the same as Microsoft 365 Cowork, which is a different product from Microsoft and has no role in AI Power Tools. See Capabilities and surfaces.

Claude Desktop. The Claude desktop application. Its chat reads skills from plugins only. It hosts the Claude Desktop extension, which supplies the server to its chat, to Claude Cowork and to the Code tab.

Cloud connection. A way EA reaches a shared repository: EA connects through Pro Cloud Server instead of connecting to the repository database directly. The repository does not necessarily run in the cloud. For AI Power Tools a cloud connection behaves the same as a server connection: EA opens the connection, and the AI Power Tools MCP Server for Sparx EA attaches to EA either way. See Supported repositories.

Code engine. A Python program bundled with a skill that computes a fixed answer on the machine where EA runs, such as the diagram advisor, the composition engine, the census, the inclusion choice, the governance gate, the Parquet and Power BI project generators and ea-com scripts. An engine gives the same result every time and uses far fewer tokens than the assistant working the same thing out. It needs local code execution and Python 3, plus pyyaml or pyarrow for some engines, and pywin32 for scripts the AI writes against EA. Building, checking and refreshing the business layer is done by the server and is not an engine. See How code engines work.

Composite diagram. A child diagram that opens when you double-click an element, so the element is its own drill-in. Two stored facts make it: the child diagram's parent is the element, and the element carries the composite flag. set_composite_diagram writes both and find_composite_diagram_mismatches audits them. See Composite and navigation diagrams.

Composition engine. The deterministic code engine of ea-diagram-composition: layout grammars that turn content into coordinates by arithmetic, composition modules, notation bindings and verification modules. See ea-diagram-composition.

Computer use. The capability that lets the assistant see and operate desktop applications: EA's own window and Power BI Desktop. It starts EA and opens a model, reads and dismisses EA's blocking dialogs, looks at the real EA window and does tasks that exist only in EA's user interface. Without it, the assistant hands you those steps, and for a blocking dialog it asks you for a screenshot. See Capabilities and surfaces.

Connector. EA's name for a relationship between two elements. It has a type and optionally a stereotype, and can carry tagged values. A connector shown on a diagram can be drawn by EA without being stored on that diagram. See Connectors.

Count back. A verification method: count the result with a query scoped to the same subtree and compare it with the plan or the source. It proves a number, not correctness.

Custom Style. Per-placement shape, opacity, rotation and similar settings stored in a diagram. It works only on placements EA created itself, not on placements added through the API. See Custom style.

D

Descriptions file. A file you approve, kept beside the reporting profile, that holds your own descriptions of business tables, columns and measures, such as what "retired" means in a lifecycle column. The Power BI generator uses it before the technology's notes and its own templates, so Microsoft 365 Copilot reads your business terms from the published model. See ea-power-bi.

Deterministic. One of the four labels. Given the same inputs and the same model state, it produces the same result, with no AI judgment involved. For anything that reads Enterprise Architect, the model state is part of the input, so the same query on a changed model returning a different answer is still deterministic. See Where the AI decides and where AI Power Tools decides.

Diagnostic report. A Markdown file the server writes when diagnostics are on and a call fails, errors or runs slowly. Nothing is sent automatically. Distinct from the narrative report that ea-diagnostic writes. See When to send a diagnostic report.

Drift (declared versus observed). The difference between what a technology declares (stereotypes, connector stereotypes, tags, enumerations) and what the model actually uses. The census reports it as findings.

E

EA (Enterprise Architect). Sparx Systems' modeling application, which AI Power Tools automates.

Embedded technology. A technology stored inside the model file, so it travels with the model. EA keeps it at Location: Project or at Location: Model. See Location (Project / Model) and Technology states.

F

Fidelity check. A comparison of a composed diagram with a reference, such as a picture, that reports differences. See Verification modules.

G

Governance gate. A step before a reporting build. It lists the in-scope elements with an empty stereotype, which will be in no business table, with the elements that carry governance tagged values first. Each element gets one of three outcomes: a suggested stereotype backed by evidence, several candidates with nothing to tell them apart (you choose), or no stereotype the technology allows for that element type. You decide per element. Applying a stereotype is a model write, taken behind a baseline. See ea-reporting-database.

GUID. EA's stable identifier for an object. Element and package GUIDs are braced. Baseline comparisons use a second, underscored form that is not interchangeable.

H

Hand-authored MDG. A technology written directly as XML, with the file as the source of truth. Contrast with an MDG built from a source model. See ea-mdg-author.

I

Inclusion choice. A decision you make once for a reporting profile, shown with counts and saved in it, about what the business layer includes beyond the technology. There are three answers: only what the technology defines; everything in the repository, with a table per observed stereotype, or per element type where there is none; or gap by gap, where each gap is included, excluded, or sent to the modeling-language skills to be declared. See ea-reporting-database.

Installed technology. A technology whose MDG file is registered with EA at the application level. One of four distinct states. See Technology states.

J

Jet. The Microsoft Access database engine behind the older .eap and .eapx local file formats. SQL written for them uses its dialect.

K

Kept extract. The rows the last reporting-database or Parquet build loaded, after the left-out packages were cut, saved in one file beside the reporting profile with the scope, counts, flags and the technology's definition. Each build replaces it, so only the last run is kept, and left-out content is never written to it. It holds included content, so treat it like the reporting data. Replay rebuilds from it. Every build also writes a definition file beside the profile that lists each table with its columns, types and row counts, which the Power BI generator reads in place of a database connection.

L

Layout grammar. A rule for turning content with a known shape into geometry, such as layered bands, lanes, nested grid, radial, matrix, chevron stack or treemap. The engine uses eleven: nine are composed by arithmetic, graph diagrams use EA's own layout and are tidied afterward, and EA-semantic diagrams cannot be produced by composing coordinates. See Layout grammars.

Linter (diagram). The module that reports overlap, alignment and fit defects in a composed diagram by rule. A clean report means no known defect, not that the layout was right.

Loaded technology. A technology live in the running EA session. EA reports a technology disabled in Manage Technologies as loaded too, so loaded does not mean enabled, and AI Power Tools reads such a technology as registered but not enabled. See Technology states.

Local code execution. The capability that lets the assistant run the Python code engines bundled with the skills, on the machine where EA runs. The diagram advisor, composition engine, census, inclusion choice, governance gate, Parquet and Power BI project generators and ea-com scripts depend on it. Without it, engine-based skills either fall back to the assistant doing the work itself, at higher token cost and lower consistency, or report that they cannot run. The server's operations, including building and refreshing the business layer, do not depend on it. See Capabilities and surfaces.

Location (Project / Model). Where Manage Technologies says a technology is stored. Location: Model holds the whole technology file, toolbox pages included. Location: Project holds profiles and diagram types but no toolbox pages. Either travels with the model. A file name in its place means a technology file on the workstation. When the model holds one id at both Locations, EA lists both and answers from the Project copy, so it shadows the Model copy. get_embedded_mdgs, get_mdg_from_runtime and install_mdg read both. See Modeling languages.

M

Manage Technologies. EA's dialog (Specialize, Technologies, Manage Technology) that shows which technologies are enabled and where each comes from. It is the cross-check when the server's lists and EA disagree.

Marketplace (plugin). A catalog a Claude client adds so it can install and update plugins. The claude.ai marketplace does not sync on its own: you sync it under Manage marketplaces and restart Claude Desktop. An uploaded plugin file never updates itself. See Install from the marketplace.

MCP (Model Context Protocol). The protocol by which an AI client calls tools on a server. The AI Power Tools MCP Server for Sparx EA uses it over standard input and output and opens no network port.

MCP client. The application that hosts the AI and calls the server: Claude Code, Claude Desktop, Claude Cowork or VS Code with GitHub Copilot. See Supported clients.

MCP server. The AI Power Tools MCP Server for Sparx EA: the Windows program that exposes EA to an AI client through six tools plus ping. See The AI Power Tools MCP Server for Sparx EA.

MDG Technology. Sparx's packaging of a modeling language for EA: a profile of stereotypes and tagged values, toolbox pages, diagram types and optional Quick Linker rules, in one XML file. Its id is limited to 12 characters. See ea_mdg and Modeling languages.

Modal dialog. A window EA raises that blocks its interface and holds the connection until a person dismisses it. It is the usual cause of a call that appears to hang. With computer use the assistant reads it, corrects the call and dismisses it; without computer use you paste a screenshot of it into the conversation, then dismiss it.

Model pattern. A reference model EA installs as a file under its ModelPatterns folder, such as a cloud reference diagram. It can be imported as a starting point, and the vendor icon libraries are read from the same place.

N

Navigation diagram. A diagram built so each element drills into a child diagram, normally a hierarchy of capabilities. It is made of composite diagrams. See ea-navigation-diagrams.

O

Operation. A named action inside a tool. AI Power Tools has 149 across six tools. In UML, "operation" also means a method of a class, and create_operation creates one. See How tools and operations work.

Orphan element. An element with no connectors, and nothing more. It can still be drawn on a diagram, contain others or anchor a composite diagram, so it is not a delete list.

P

Parquet. A columnar data file format, and one of the three delivery ways for the business layer: one Parquet file per table in a local folder, written with the pyarrow package and with no database. The generated Power BI project imports the files from an absolute path.

.pbip. A Power BI project saved as text files. The skill generates one, and it does not load data on open. See ea-power-bi.

Plugin. A Claude package that carries the skills and no server. Installed once on your Claude account, it reaches Claude Desktop chat, Claude Cowork and the Code tab. Its skills are named ai-power-tools:<name>. See Claude plugin.

Presentation profile. A named setting in a binding that says what a view shows: detail, review or executive.

Pro Cloud Server. Sparx's middleware that lets Enterprise Architect reach a repository database over HTTPS instead of through a direct database connection. A repository reached this way is a cloud connection. It is not a server in any special sense and does not necessarily run in the cloud. For AI Power Tools a cloud connection behaves the same as a server connection.

Profile (UML profile). The part of a technology that defines stereotypes and their tagged values, as extensions of metaclasses. publish_package_as_profile exports a profile package as a versioned profile. See Publishing a profile.

Profile package. A package in a model that holds the source definition of a profile. Check for one before choosing how to build an MDG. See ea-mdg-model-build.

Prolaborate. A separate Sparx product, the hosted collaboration layer over an EA repository: reviews, dashboards, matrices and impact analysis. The Prolaborate skills drive its web interface in your signed-in browser session. See Prolaborate.

Q

Quick Linker. EA's feature that suggests and creates relationships when you drag from an element. A technology can define Quick Linker rules. A constraint must be qualified as <TechID>::<Stereotype>, and verification needs the interface.

R

Read-only mode. A behavior of the AI Power Tools MCP Server for Sparx EA that refuses operations that modify the repository, including diagram layout operations, decided by operation name. It is the Read-only mode option on the Claude Desktop extension's Configure screen (off by default), and it sets EA_MCP_READONLY in the server's environment. It is not a security boundary.

Reference data. EA's stored definitions of tagged value types (enumerations and their values). write_mdg_xml ships none, so enumerations can appear as free text.

Refresh. Bringing the business layer up to date. One call starts it in the background and returns a run id, and a second call polls it until it reports succeeded or failed. It runs whichever delivery way the profile names. A scheduled AI agent can start a refresh. A run that fails, or that the reporting preflight stops, leaves the previous data live.

Registered technology. A technology EA knows about in its lists, with a Location and an enabled flag. A registered technology can be disabled. See Technology states.

Replay. Rebuilding a separate reporting database, or the rows for Parquet files, from the kept extract with no EA connection. It reproduces that build's instant. It is not a refresh, and it does not apply to views in the EA database.

Reporting database. A separate database, on SQL Server or a SQL database in Microsoft Fabric, that holds a scoped copy of EA's tables in EA's own shapes, with the business layer built over it as views. It is one of the three delivery ways, needs no direct access to the EA database, is a snapshot rebuilt in full on every refresh, and is read by Power BI with Import.

Reporting preflight. A dry run before a reporting-database or Parquet build. It runs the extract and the cut of the left-out packages and writes nothing. It reports the rows that would be kept, the links that cross into a left-out area and the "(excluded)" stand-ins they need, the defects that block the build, and the flags to show you before you continue. It is not the session preflight.

Reporting profile. A file you keep that holds the technology the business layer follows; the scope, as root packages to include and packages to leave out, both by identifier so that a renamed or moved package keeps its setting; the inclusion choice and a record of it; and where the output goes. Every build, check and refresh reuses it.

Repository. The EA model store, reached as a local file (.qea or .qeax), a server connection or a cloud connection. See Supported repositories.

Rule-based. One of the four labels. AI Power Tools applies fixed rules to make a judgment, such as recommending, ranking, classifying or flagging. Repeatable, but it can be wrong.

Ruleset. A YAML file of conformance rules, with selectors and conditions, that ea_validate audit runs. One ships, for ArchiMate 3.1. See ea-validation.

S

Selector (ruleset). The part of a rule that chooses what it applies to, such as elements of a stereotype or connectors of a stereotype. It matches an exact stored string.

Semantic model (Power BI). The data model Power BI holds: tables, relationships and measures. The skill writes it as text files in TMDL format, and Power BI rejects the whole project on one violation.

Server connection. A way EA reaches a shared repository: EA connects directly to a repository held in a database. Contrast with a local file (.qea or .qeax) and a cloud connection. The AI Power Tools MCP Server for Sparx EA reaches the repository through EA and does not rewrite SQL for it. See Supported repositories.

Session preflight. The short checks a session runs first: ping, get_repository_info, assess_mdg_situation and a confirmation before the first write. For Prolaborate, five read-only checks of sign-in, token, identity, version and repositories. It is not the reporting preflight. See ea-start-here.

Skill. A folder containing a SKILL.md that tells the AI how to do a kind of work: the order to do things in, the mistakes to avoid and the checks that prove a result. Skills are text the AI follows, not code that runs by itself. AI Power Tools has 24. See How skills work.

Snapshot. A record of a repository at one instant. A separate reporting database and Parquet files are snapshots. Views in the EA database are live.

SQLite. The database engine behind .qea and .qeax local files. add_image needs a local file.

Stereotype. A name that extends an element's or connector's meaning, such as WBA::WBABusinessApplication. It can be profile-bound or ad hoc.

Surface. The application you work with the assistant in, such as Claude Code, Claude Cowork, Claude Desktop chat or VS Code with GitHub Copilot. Surfaces differ in which capabilities they give the assistant: the MCP server, computer use, local code execution and browser control. They gain capabilities over time. See Capabilities and surfaces.

T

Table Directory. The index of the business tables: one row per table with its name and kind, its stereotype, alias and element type, the tables at each end for a connector table, whether the technology declares the table, and the technology's description.

Tagged value. A named property that a stereotype defines and an element, connector, attribute or operation carries, such as a criticality or an owner. See Tagged values.

Technology (installed, embedded, registered, loaded). Distinct states of an MDG Technology, and none implies another. Installed: the MDG file is registered at the application level. Embedded: the technology is stored in the model file, at Location: Project or Location: Model. Registered: EA's lists know of it, with a Location and an enabled flag. Loaded: it is live in this EA session. EA reports a technology disabled in Manage Technologies as loaded, so a technology can be loaded and not enabled. A true result from an import call is not proof, and a delete takes effect only after a restart. See ea_mdg.

Technology id. The short identifier EA registers for an MDG, limited to 12 characters, and the prefix of qualified stereotypes. Westbrook uses WBA.

Toolbox. The page of stereotype tools EA shows for a diagram type. Its page name must equal the toolbox name in the diagram profile. See ea-mdg-deploy.

V

Vendor icon library. The AWS, Azure, Google and data-platform icon sets that EA installs as model pattern files. find_icon and set_element_icon read them from the local EA install. See Vendor icon libraries.

Verify (a diagram). verify_diagram reloads a diagram, renders it, and returns the canvas extent, every placed object with its rectangle and stacking order, and every link with whether a stored record backs it. See Verifying a diagram.

Viewpoint. A named selection of what a diagram shows for a stakeholder. Only the ArchiMate 3 binding declares viewpoints, and the diagram advisor ranks them. See ea-diagram-advisor.

.vsix. The VS Code extension package. It bundles the server and adds the released-skills marketplace to VS Code; the skills are a separate plugin. See VS Code extension.

W

Westbrook Bank. The fictitious reference organization used for every example in this reference: technology id WBA, display name WestbrookBankArchitecture, stereotype WBA::WBABusinessApplication, model WestbrookBank.qea. It is an illustration. No step here assumes you have it.

X

XMI. The XML Metadata Interchange format for model interchange. export_xmi supports versions 1.0, 1.1, 1.2, 2.1 (the default), 2.4.1 and 2.5.1, and checks the file holds the package's elements. XMI is not a backup. See Import and export.

Z

Z-order. The stacking number of a diagram placement. A lower number is nearer the front, and a filled container needs a higher number than its contents.