Draw a cloud architecture diagram with vendor icons

Goal

Draw AWS, Azure or Google Cloud elements as their vendor icons instead of plain boxes. For a cloud or solution architect. You end with a diagram whose placed elements show the right vendor glyphs, chosen deliberately, with the images held in the model's image library and the result confirmed by a render.

Where it runs

This scenario needs the MCP server. Computer use covers its UI steps.

Before you start

• The start-of-session check done.
• A diagram with the elements already placed (Start from a model pattern, Compose a diagram, or your own).
• To load an icon that the model does not already hold: a local file (.qea or .qeax), because EA's automation interface has no way to reach the image library. On a server connection or a cloud connection, load the artwork through EA's Image Manager, a UI step that is AI with computer use; otherwise you, and the operations below then reuse it.
• EA installed with its ModelPatterns folder. If AI Power Tools cannot find the folder, it reports that the icon libraries were not found, and the EA_MODEL_PATTERNS setting points it at the folder.

EA can draw an element as a picture. The artwork lives in the model's image library, and the assignment is stored on the placement, so other diagrams that show the same element are untouched. EA ships AWS, Azure and Google Cloud icon sets, about 5,200 icons on EA 17.1, as model-pattern files in the EA install folder. AI Power Tools reads these libraries in place from your own EA install. Nothing is downloaded and no artwork ships with the server.

Steps

1. AI with computer use; otherwise you If EA is not running or the model is not open, start EA and open it. AI Power Tools ea-start-here runs the start-of-session check and confirms the project type.
2. You Decide the concept for each element. ea-diagram-composition and ea-modeling propose them. The aim is the concept, not the caption. A diagram's captions are not icon names. Amazon S3 output bucket wants Amazon Simple Storage Service (S3) or Bucket with objects. Function App wants Function Apps, and SQL Database wants SQL Databases. Containers and labels such as a region or a tier never had an icon. Some glyphs are not in the libraries at all: there is no plain AWS key. Say so rather than substituting a near miss.
3. AI Power Tools ea-modeling shows what the model already holds in its image library (list_images).
4. You ea-modeling searches the shipped libraries (find_icon, rule-based), and you choose among the candidates. Matches are ranked exact first, then normalized (ignoring case, punctuation, a file extension, a leading vendor word such as Amazon or Azure, and a trailing plural), then every-word, then partial. Within each kind the provider's newest vintage comes first, so the same glyph often appears once per vintage. The provider can be narrowed to AWS, Azure, Google or Data platform, and the default is ten results. Each match shows its provider, vintage, package, how it matched, and whether the model already holds it. Choose among the candidates, and show them to the user when more than one fits. Prefer the newest vintage. Set only the icons the diagram uses: the full set is tens of megabytes.
5. AI Power Tools ea-modeling assigns the icon to each placement (set_element_icon). It reads the icon from your install, reuses the model's image if one exists under that name, or loads it first, then sets it on the placement. The stored image is named <provider> <vintage> - <name>, for example AWS v19 - AWS Lambda. Where one library has same-named icons whose pictures differ, each name carries a short id in brackets. Refusals come back unchanged, for example when the icon must be loaded and the model is not a local file, or when the element is not on the diagram. An image added before a refusal stays in the library, and the next request reuses it.
6. You Look at every icon in the render (verify_diagram, from ea-diagram-composition). A success reply and a verified flag mean the setting was read back out of EA after a reload. They do not mean EA rendered the picture.
7. AI Power Tools ea-modeling corrects: it removes an assignment and assigns another (clear_element_image), then repeats steps 4 to 6.

If you already have icon files on disk (PNG, EMF or WMF, up to 1 MB each), ea-modeling can load each (add_image) and assign it (set_element_image). Loading is idempotent on content: adding artwork the model holds returns the existing image. It refuses JPEG, GIF and BMP with a message naming the conversion, because EA re-encodes raster artwork to PNG and storing other bytes under the bitmap type would draw a blank element and report nothing. Assigning by name is refused when the name matches more than one image, because image names are not unique and are case-insensitive. Assigning an image id with no image behind it makes EA draw the element normally with no complaint, so AI Power Tools checks the id before it writes. Clearing stops drawing the image and leaves the image in the library, so other placements that use it are unaffected.

Prompts to use

Opening

Follow-ups

A case in which the libraries have no match.

Loading through Image Manager.

Loading icon files from disk.

Recommended models

Ratings are guidance based on the kind of reasoning the skill needs, not measured benchmarks.

Checking the result

• Accurate. In the render, each assigned element shows the glyph for its concept. The image list shows the images added, with their <provider> <vintage> - <name> names. The two checks of each assignment, the stored setting and the picture, agree.
• Complete. Every element that should carry an icon has one, or the report says why not. Containers, labels and concepts with no icon in the libraries are named.
• Nothing unintended changed. Only the chosen placements on this diagram changed. Other diagrams that show the same elements are unchanged. The image library holds only the images you meant to add.

Variations

• Start from a shipped pattern. Start from a model pattern imports a pattern and then runs these steps.
• Data-platform icons. Set the provider to Data platform in the search.
• A model on a server connection or a cloud connection. Load artwork through EA's Image Manager, a UI step that is AI with computer use; otherwise you, under the name AI Power Tools would give it, and it reuses it.
• Replace an icon. Clear the assignment, then assign another.

Limitations

• This scenario gives partial results. Azure comes out well, AWS and Google Cloud partly, and captions are not drawn.
• Captions are not drawn under image elements in the exported PNG, whether or not the caption setting is on. This is reported across the patterns tried.
• Fidelity by provider, as reported. In the patterns tried, Azure icons matched in kind and color, and the only gap was captions. AWS glyphs were right and colored where the reference uses dark tiles or outlines. Some Google Cloud elements render as text with no hexagon glyph.
• A glyph can differ from its caption. One icon named for a key is a document with a key and a padlock, not a plain key.
• Vendor logos and corner badges are not covered. A badge that is part of a container cannot take an image without replacing the container. A logo element takes an image, and the library's version may be a bare mark that EA stretches into the slot.
• A success reply is a stored setting, not a rendered picture.
• Loading needs a local file. Only a local file (.qea or .qeax) accepts the server's loading operation. On a server connection or a cloud connection, load through EA's Image Manager.
• A library icon whose picture is larger than 1 MB is refused. That is also the size limit for icon files you load yourself. In the EA 17.1 libraries examined, the largest icon is about 158 KB.
• The libraries come from your EA install. A different build can ship different files. List the folder, and expect an icon-libraries-not-found message where it is missing.
• The icons are the vendors' marks, already on the machine through the EA license. Nothing is downloaded or shipped by the server.