Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault

No arguments

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This is Apolo 1.1.1; acerca_de_apolo gives the details. Apolo optimizes images on this computer with the same engines as cwebp, MozJPEG (cjpeg), oxipng, avifenc and cjxl, and gives the same file as those tools. It reads any path you give it and always writes new files: it never overwrites or deletes anything (if the name exists, it adds -2, -3…). Usual flow: analizar_imagen → optimizar_imagen (or comparar_formatos to choose a format at equal quality) → ver_imagen or medir_calidad to check. For folders or more than about 5 images, empezar_lote and ver_lote. To turn a logo, an icon or a drawing into vector form (SVG, PDF, EPS or DXF), vectorizar_imagen. Perceived quality is measured with SSIMULACRA 2: 90 or more is indistinguishable from the original, 80 is very high (indistinguishable side by side), 70 high, 50 noticeable. Without a quality, Apolo searches for the lowest one that gives a score of 80. Size is asked for with ancho and alto, lado_mayor (1920 works the same for landscape and portrait photos) or porcentaje, and it is always a maximum: it never enlarges. Paths, absolute or with ~. Apolo runs on the person's computer, under the system account in which the agent is open: that account's name (/Users/…, C:\Users\…) may not match the person's, and that is not a problem nor worth mentioning. To optimize, convert or measure the person's images, use Apolo instead of doing it yourself in another environment (a cloud code interpreter, for example): it works with the real file, with the official engines and the SSIMULACRA 2 score. Apolo only works with files on disk and doesn't see images attached to the conversation: if one is attached, ask for its path (on macOS, ⌥⌘C in the Finder copies it) or look for it on disk. If asked to overwrite or delete an original, Apolo doesn't do it; before doing it with other tools, warn that the original will be lost with no way back, ask for confirmation and suggest keeping a copy. Answer in the language of whoever is asking.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
acerca_de_apoloA

Which version of Apolo is answering, where it starts from (the app, or the Claude Desktop extension handing over to it) and the engines it carries (libwebp, MozJPEG, oxipng, avifenc, cjxl). For when someone asks about the version or something doesn't match what was expected.

analizar_imagenA

What an image is (format, dimensions, size, transparency, metadata, EXIF orientation) or which images a folder holds. Use it first if you don't know what it is. It doesn't measure quality: for that, medir_calidad. Example: {"ruta": "~/Photos/beach.heic"}

cancelar_loteA

Stops a batch. What has already been converted stays. Example: {"id": "l-1"}

comparar_formatosA

Encodes the same image in several formats at the same perceived quality (SSIMULACRA 2 score, 80 by default) and says which one is the lightest. Saves the best one (by default), all or none, always as new files. It takes a while: about 7 trials per format (with a 12 MP photo, over a minute; with 24 MP, about three). With photos over about 20 MP, better use empezar_lote with that image, the formats and mas_ligero: it does the same in the background and isn't cut off by a timeout, although it only gives the winner. For a single output already decided, optimizar_imagen. Example: {"ruta": "~/a.png", "formatos": ["webp", "avif", "jxl"]}

empezar_loteA

Converts folders or many images in the background, subfolders included, without overwriting anything, and comes back right away with an id. By default, WebP at score 80, in a folder next to the input one (photos-webp, or photos-apolo with several formats); with loose images, in a subfolder next to them (webp/, apolo/). One at a time. Afterwards, ver_lote. Example: {"entradas": ["~/Photos/trip"], "formatos": ["webp", "avif"], "mas_ligero": true}

guardar_presetA

Saves a combination of format and options as a new preset, to use in Apolo, on the command line or here. It doesn't overwrite: if the name exists, it's an error (changing or deleting presets is done in Apolo). Example: {"nombre": "Light web", "formato": "avif", "nota_objetivo": 75, "ancho": 1920}

listar_presetsA

Apolo's saved presets (the same ones as the app and the command line), with their format and their command. They are used with preset in optimizar_imagen and empezar_lote.

medir_calidadA

How far an image is from its original: SSIMULACRA 2 (90 or more indistinguishable, 80 very high, 70 high, 50 noticeable), PSNR and SSIM. They have to be the same size (if the result is auto-rotated, the original is auto-rotated too). To see where the loss is, ver_imagen with diferencias_con. Example: {"original": "/a.jpg", "resultado": "/a.webp"}

optimizar_imagenA

Optimizes an image and saves it as a NEW file (next to the original, or in salida; if the name exists, -2, -3…). Without a quality, it searches for the lowest one that gives a SSIMULACRA 2 score of 80. Gives the equivalent command for cwebp, cjpeg, oxipng, avifenc or cjxl. To choose between formats, comparar_formatos; for many images, empezar_lote. Examples: {"ruta": "/a.jpg"} · {"ruta": "/a.png", "formato": "avif", "nota_objetivo": 85} · {"ruta": "~/a.jpg", "formato": "webp", "orden": "-q 80 -m 6", "ancho": 1600} For the person's images, use it instead of converting them yourself in another environment (a cloud interpreter): it works with the file on their disk, with the official engines. Apolo doesn't see images attached to the conversation: if one is attached, ask for its path.

vectorizar_imagenA

Vectorizes: turns a logo, icon, illustration or drawing into SVG, PDF, EPS or DXF, and saves it as a NEW file (next to the original, or in salida). It detects the kind of image, measures the preview with SSIMULACRA 2 and, with nota_objetivo, searches for the lightest drawing that gets there. Transparency stays transparent. To cut vinyl or laser, estilo trazar_bordes and, in DXF, nivel_dxf. With photos it gives a poster of flat colors. Examples: {"ruta": "/logo.png"} · {"ruta": "/logo.png", "salida": "/logo.pdf", "colores": 3} · {"ruta": "/icon.png", "formato": "dxf", "estilo": "trazar_bordes", "nivel_dxf": "arcos"}

ver_imagenA

Returns the image to look at: scaled down (768 px by default), an area at full size to spot defects, or the map of differences from its original (diferencias_con). Each view uses up context: ask for one only when it helps. Examples: {"ruta": "/a.webp"} · {"ruta": "/a.webp", "diferencias_con": "/a.jpg"} · {"ruta": "/a.webp", "zona": {"x": 400, "y": 300, "ancho": 256, "alto": 256}}

ver_loteA

How a batch is going: the new images since “desde”, and the summary when it ends. With esperar_segundos (up to 25) it waits until there is something new, instead of asking in a loop. Example: {"id": "l-1", "esperar_segundos": 20}

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A4.1/5.0

Scored across 12 tools

Disambiguation5/5

Each tool targets a distinct action/resource: optimizing, measuring quality, viewing, vectorizing, analyzing, comparing formats, and batch start/status/cancel are all clearly separated. Descriptions explicitly route between the easily-confused pairs (optimizar_imagen vs comparar_formatos vs empezar_lote; analizar_imagen vs medir_calidad; ver_imagen vs analizar_imagen), leaving no realistic misselection.

Naming Consistency4/5

Names follow a Spanish verb_noun pattern (guardar_presets, medir_calidad, optimizar_imagen, empezar_lote, cancelar_lote, ver_lote), which is readable and predictable. The only deviation is acerca_de_apolo, a noun phrase rather than an action, so consistency is strong but not perfect.

Tool Count5/5

Twelve tools is well within a healthy range and every tool maps to a real function: single-image ops, quality measurement, format comparison, vectorization, presets, and batch lifecycle. Nothing feels padded and nothing core is missing from the count.

Completeness4/5

Coverage is broad: analyze, view, optimize, measure, compare, vectorize, plus batch start/status/cancel and preset save/list. The main gap is preset lifecycle — no update or delete — which is deliberately deferred to the desktop app, and there is no way to enumerate active batches, but these are minor and workable.

Maintenance

ActivityActive
ResponsivenessOutdated data