Tintloom
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| FAL_KEY | No | Provider key for fal.ai. | |
| ARK_API_KEY | No | Provider key for BytePlus ModelArk. | |
| GEMINI_API_KEY | No | Provider key for Google Gemini. | |
| OPENAI_API_KEY | No | Provider key for OpenAI. | |
| TINTLOOM_PUBLIC_URL | No | The public URL for the HTTP server (required when running as an HTTP server). | |
| TINTLOOM_SERVER_TOKEN | No | A long random string used as the token for the HTTP server connector URL (required when running as an HTTP server). |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Image generation and editing. Use generate_image whenever the user asks to create, draw, render or edit an image. Each result says whether the user can already see the image.
Editing: every result lists an id per image. When the user asks to change an image, in a chat message or through the image card, call generate_image again with images: [{ id }] of the image they mean (the latest one unless they say otherwise) and a prompt that says what changes and what stays. Never start over from text when an image exists.
Messages from the image card say "this image" or "the selected area" and give no id: call generate_image with selection: true and leave out images, mask and erase. The server knows which image is open and what the user painted, and whether it is an erase. For a selected area, write the change as the prompt and name the object it is about if you can tell ("make the kite red"); it keeps its shape and position.
One part only, without the card: add mask or edit_regions; everything outside is kept exactly. Name the object inside the area in the prompt ("make the kite red", not "make it red"). To remove something: the same, plus erase: true.
Remove the background: provider openai, model gpt-image-1.5, background transparent, fidelity high. New shape: aspect, with a prompt to keep the scene and extend it. Variations: count. Several different images at once: prompts, one prompt each, shown as one gallery.
Make one image per request unless asked for more: each image costs the user money.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
| resources | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| generate_imageA | Create an image from a text prompt, or edit or combine existing images when |
| list_image_providersA | List the image providers and models that can be used, what each can do, and which have a key set. |
| get_upload_urlA | Answers how to give this server an image file: it runs on the user's machine and reads files directly, so pass the file's path in |
| save_maskA | Used by the image card to store the area the user painted. Not useful to call directly. |
| save_selectionB | Used by the image card to say which image the user has open and what they painted on it. Not useful to call directly. |
| get_progressC | Used by the image card to show images as they finish. Not useful to call directly. |
| save_layer_arrangementA | Used by the image card to keep what the user moved and hid in a layer set, so it is there after a reload (and on the dashboard). Not useful to call directly. |
| get_layer_setA | Used by the image card to reopen a split-into-layers result after the chat is reloaded. Not useful to call directly. |
| get_generated_imageC | Used by the image card to load an image by id. Not useful to call directly. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| Image card |
TDQS
Scored across 9 tools
The main agent-facing tools (generate_image, list_image_providers, get_upload_url) are reasonably distinct, but six tools are explicitly internal image-card helpers that agents should not call directly. Some boundaries are still fuzzy, such as save_mask vs. save_selection and get_generated_image vs. get_layer_set, even though the descriptions clarify them.
All tool names use consistent snake_case with clear verb_noun phrasing (get_*, save_*, generate_image, list_image_providers). There are no mixed naming conventions or vague standalone verbs.
Nine tools is not excessive in raw count, but the surface is inflated by six image-card UI helpers that are not useful for an agent to call directly. The effective agent-facing surface is only about three tools, making the set feel heavier and less focused than it should be.
The core image-generation workflow is well covered: generating, editing, combining, splitting into layers, multi-prompt galleries, provider listing, and local file-path guidance. Minor gaps remain around image history/listing, cancellation, or explicit deletion, but the main lifecycle is usable.