Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@ImgLume MCP Browse popular ImgLume inspiration and show me three different visual directions."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An open-source local MCP server that connects AI clients to ImgLume for visual inspiration, prompt improvement, image generation and editing, and video generation.
The tools are flexible building blocks. Search the gallery when inspiration is useful, improve a short idea when needed, or generate directly when the user already knows what they want.
MCP host (Codex, Claude Code, Cursor, ...)
-> local ImgLume MCP server (this repository)
-> ImgLume API
-> generated files saved on your computerThis repository contains only the open-source MCP server. The ImgLume website and API backend are not included.
Gallery prompts and model information are read from ImgLume's live catalog. New gallery entries, model options, and image or video credit costs can therefore become available without an MCP update. An MCP update is needed only when the tool or API contract changes.
Related MCP server: fluxlora-mcp
Every asset below came from one production run through the public MCP tools.
The fictional LUME NO. 7 product was created first, combined with a fashion
model reference, and then animated from the saved local campaign image.
Create the product | Build a model campaign | Animate the campaign |
|
|
|
Watch the 4-second generated video.
These are ordinary generate_image and generate_video calls rather than
hard-coded workflows. They show how the same tools can handle several
constraints at once.
The base bottle was used in four parallel image requests: fashion campaign, midnight neon, moonlit botanicals, and minimal editorial. The bottle geometry, purple-to-amber liquid, black cap, and gold label remain recognizable while the art direction changes.
Left to right: the product reference, the model reference, and a new campaign. The request preserved the model's identity, green dress, bottle design, and label while moving both subjects into a warm gold studio.
This 9:16 poster required exactly two headline lines:
MIDNIGHT BLOOM and LUME NO. 7, while preserving the three-line product
label.
Frames at 0.05, 1.30, 2.60, and 3.85 seconds show the same face, hairstyle, dress, hand pose, bottle, and label through a slow camera push-in and subtle subject motion.
Watch the full 4-second continuity test.
Node.js 20 or newer
An MCP client that supports local stdio servers
An ImgLume account and API key only when generating images or videos
Gallery search, inspiration details, prompt improvement, and model discovery do not require an API key.
Add ImgLume to your MCP client:
{
"mcpServers": {
"imglume": {
"command": "npx",
"args": ["-y", "github:Hanyuyu/imglume-mcp"]
}
}
}Restart the client, then try:
Browse popular ImgLume inspiration and show me three different visual directions.
To enable image and video generation:
Create an ImgLume API key. ImgLume will guide signed-out users through login first.
Add the key to the server's environment:
{
"mcpServers": {
"imglume": {
"command": "npx",
"args": ["-y", "github:Hanyuyu/imglume-mcp"],
"env": {
"IMGLUME_API_KEY": "imglume_your_key"
}
}
}
}Keep the key in your private, local MCP configuration. Never commit it to a repository.
codex mcp add imglume -- npx -y github:Hanyuyu/imglume-mcpAdd --env IMGLUME_API_KEY=imglume_your_key before -- when generation is
needed.
claude mcp add --scope user imglume -- npx -y github:Hanyuyu/imglume-mcpAdd -e IMGLUME_API_KEY=imglume_your_key before -- when generation is
needed.
Save the JSON configuration above using Cursor's MCP configuration:
~/.cursor/mcp.json for all projects
.cursor/mcp.json for one project
Restart Cursor after saving the file.
Use natural-language requests. The host chooses the relevant tool.
Search ImgLume for poster inspiration and show me five options.
Use short, broad search terms such as poster, product, portrait, anime,
or cinematic. Leave the search empty to browse popular work.
Improve this into a production-ready image prompt: a glass of coffee on a snowy mountain, editorial style.
Generate a vertical luxury campaign for a fictional LUME NO. 7 perfume: an adult East Asian fashion model in an emerald one-shoulder dress, warm studio lighting, and the product label clearly readable.
Use
/absolute/path/to/product.pngand/absolute/path/to/model.pngas references. Preserve the model's identity, dress, bottle shape, and label; place them together in a warm gold studio.
Turn
/absolute/path/to/campaign.pnginto a four-second video. Add a subtle turn, blink, hair movement, and slow camera push-in while keeping the face, hands, bottle, and label stable.
Preview rendering depends on the MCP host. Hosts without inline media support may show a URL instead of the image or video itself.
Tool | Purpose | API key |
| Search the live gallery for visual references and prompt examples | No |
| Retrieve the full prompt, media, recommendations, and attribution | No |
| Ask the current host model to improve a short visual brief | No |
| Read current models, options, defaults, and live credit costs | No |
| Generate or edit one image and save it locally | Yes |
| Generate one video and save it locally | Yes |
Each tool can be used independently. Gallery search is optional inspiration, not a required step before generation.
Use list_models when model choice, supported options, or current costs
matter. Avoid relying on a hard-coded model list.
Image costs are reported per generation. Video costs are reported per second, so multiply the selected rate by the requested duration.
Image generation and video generation consume credits from the ImgLume account associated with the API key.
Reference images may be public HTTP(S) URLs or local file paths.
Local paths should be absolute so the MCP process resolves the intended file.
JPG, PNG, and WebP reference images are supported.
Each reference image may be up to 10 MB, with at most five references per request.
The MCP validates the live model, quality, and aspect-ratio options before uploading local references. Partial uploads are removed when an upload or generation submission fails.
Video duration may be between 4 and 15 seconds. Available quality and aspect ratio values come from the live model catalog.
Generated files are saved to ~/Pictures/imglume by default.
Set IMGLUME_OUTPUT_DIR to use another save location.
Variable | Purpose | Default |
| ImgLume API key used for generation | None |
| Advanced API origin override |
|
| Where generated files are saved |
|
| Generation status polling interval |
|
| Image wait timeout |
|
| Video wait timeout |
|
Most users should leave IMGLUME_API_URL unset. It exists for ImgLume
maintainers and trusted compatible API environments; it does not provide the
private ImgLume backend for local development.
The configured API origin receives the API key, prompts, and uploaded reference
images. Only point IMGLUME_API_URL at an endpoint you trust.
search_gallery, get_inspiration, and list_models read data from the
ImgLume API.
enhance_prompt returns guidance to the current MCP host model and does not
call a separate ImgLume AI model.
Image and video generation send the prompt and any reference images to ImgLume.
Generated assets are downloaded to the configured local output directory.
This MCP server adds no analytics or telemetry.
Treat IMGLUME_API_KEY like a password. Do not publish it, commit it, or
paste it into shared logs.
See the ImgLume Privacy Policy for the service's data practices.
Confirm that Node.js 20 or newer is installed, then restart the MCP client after changing its configuration.
Use a shorter or broader term, or omit the query to browse popular inspiration.
Create a key at ImgLume API Keys, add it
as IMGLUME_API_KEY, and restart the client so it reloads the environment.
Use an absolute local path or a public URL that returns a JPG, PNG, or WebP image. Confirm that each image is no larger than 10 MB.
The MCP server waits up to 5 minutes for images and 15 minutes for videos by
default. The MCP host's own request timeout must be at least as long. Increase
the host timeout or the corresponding IMGLUME_*_TIMEOUT_MS value when needed.
The completed tool response includes the exact saved path. The default
directory is ~/Pictures/imglume; check IMGLUME_OUTPUT_DIR if it was
overridden.
For reproducible problems, open a GitHub issue without including your API key.
skills/visual-creative/SKILL.md provides
lightweight guidance for hosts that support skills. It keeps scenarios such as
product imagery, portraits, posters, characters, interiors, thumbnails, and
storyboards as examples rather than fixed workflows.
The MCP server works without the skill.
This public repository contains the MCP layer only. The private ImgLume product and API backend are not part of the contributor setup.
git clone https://github.com/Hanyuyu/imglume-mcp.git
cd imglume-mcp
pnpm install
pnpm check
pnpm buildContributors can exercise the read-only tools against the public ImgLume API. Generation tests require their own ImgLume account, API key, and credits.
Turn a short visual idea into a stronger generation prompt using the current host model. This tool returns concise guidance rather than calling another AI service.
| Name | Required | Description | Default |
|---|---|---|---|
| style | No | Optional visual direction such as realistic, anime, editorial, 3D, or watercolor. | |
| prompt | Yes | The brief idea to improve |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, but the description adds context by stating the tool uses the host model and returns guidance, not invoking external services. This clarifies the behavior beyond the annotation, though it omits details like whether the input prompt is modified or used for training.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences long, front-loaded with the core action and key differentiator. Every sentence adds value, with no redundant or filler content. Efficient and to the point.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with 2 parameters, readOnlyHint annotation, and no output schema, the description adequately covers the function and behavior. It explains the tool does not call another AI service, which is important context. However, it does not describe the return format or how the guidance is presented, leaving a minor gap in completeness.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the baseline is 3. The description mentions 'short visual idea' (maps to prompt) and 'visual direction' (maps to style), but these are paraphrases of the schema descriptions without adding new meaning or format details. No additional value over the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose: 'Turn a short visual idea into a stronger generation prompt.' It explicitly distinguishes from siblings like generate_image/generate_video by noting it 'returns concise guidance rather than calling another AI service.' This provides a specific verb and resource, fully differentiating from alternatives.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies use cases (when you have a short visual idea and want a better prompt) but does not explicitly state when to use versus alternatives or provide exclusions. Sibling tools like search_gallery or generate_image are not contrasted, leaving the agent to infer context without clear guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Generate or edit one image with ImgLume. For edits, pass the source image URL or local file path in referenceImages and keep the prompt focused on the requested change. Omit model, quality, or aspectRatio to use live ImgLume defaults.
| Name | Required | Description | Default |
|---|---|---|---|
| model | No | Optional model ID | |
| prompt | Yes | Image generation prompt | |
| quality | No | Optional quality value | |
| promptId | No | Optional ID returned by get_inspiration | |
| aspectRatio | No | Optional ratio such as 1:1, 16:9, or 9:16 | |
| referenceImages | No | Public image URLs or absolute/local image file paths |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description must carry the full burden. It discloses the tool can generate or edit, but it does not mention return format (e.g., image URL or file path), error handling, rate limits, or potential side effects. This is a significant gap for a tool that produces output.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, dense sentence with no redundancy. It front-loads the main action and immediately provides usage tips. Every phrase adds value, achieving maximum conciseness without sacrificing clarity.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity (6 parameters, no output schema, no annotations), the description is adequate for basic usage but lacks important details such as output format, error handling, or processing time. A more complete description would include what the tool returns or how to handle failures.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, but the description adds meaningful context: referenceImages should contain source file for edits, and prompt should focus on the change. It also suggests omitting certain parameters to use defaults, which enhances understanding beyond the schema definitions.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Generate or edit one image with ImgLume', specifying the verb (generate/edit) and resource (image). It distinguishes from sibling tools like generate_video (video) and search_gallery (searching), establishing a unique purpose.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides explicit guidance for editing: pass source image in referenceImages and keep prompt focused. It also advises omitting model, quality, aspectRatio to use defaults. While it doesn't explicitly state when not to use the tool, the context is clear for typical usage.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Generate one video with ImgLume. Pass a reference image URL or local path for image-to-video. Omit model, quality, or aspectRatio to use live ImgLume defaults.
| Name | Required | Description | Default |
|---|---|---|---|
| model | No | Optional video model ID | |
| prompt | Yes | Describe subject motion, camera motion, and scene changes | |
| quality | No | Optional video quality | |
| promptId | No | ||
| aspectRatio | No | Optional ratio such as 16:9 or 9:16 | |
| generateAudio | No | Whether to request generated audio | |
| durationSeconds | No | Video duration from 4 to 15 seconds | |
| referenceImages | No | Public image URLs or absolute/local image file paths |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, and the description only states it generates a video. Missing details on execution time, async behavior, destructive actions, or authentication needs.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with purpose, no extraneous information. Every word earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given 8 parameters and no output schema, the description lacks details on return values, error handling, or parameter interactions. Basic functionality is covered, but experienced users need more context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 88%, so most parameter descriptions are already present. The description adds value by noting that omitting model/quality/aspectRatio uses defaults, but does not explain advanced parameters like promptId or generateAudio.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The phrase 'Generate one video with ImgLume' clearly states the action and resource. Mentioning image-to-video distinguishes it from sibling tools like generate_image, but it could be more explicit about output type.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides implicit guidance by mentioning optional parameters and defaults, but does not explicitly state when to use this tool versus siblings or when not to use it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get the full prompt, preview media, recommendations, and attribution for one result returned by search_gallery.
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | ImgLume inspiration ID |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Description discloses key behavioral trait: it returns comprehensive data for one result. Annotations already indicate readOnlyHint, so description enhances with specific return contents.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single sentence that is front-loaded and contains all essential information. No redundancy or unnecessary words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Tool has one parameter, no output schema, and description explains what is returned. Given simplicity and no nested objects, description is fully complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Parameter 'id' is fully described in schema (ImgLume inspiration ID). Description adds context that id comes from search_gallery, but this is marginal beyond schema. Baseline 3 due to high schema coverage.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states it gets full details for a single result from search_gallery, listing specific elements (prompt, preview media, recommendations, attribution). Distinguishes from sibling search_gallery which returns a list.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Implicitly clear that this tool is used after search_gallery to retrieve detailed information for a specific result. No explicit exclusions or alternatives, but context is sufficient.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List the image and video models currently available from ImgLume, including supported quality, aspect-ratio, and live credit-cost values. Use when model choice, capability, or cost matters.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations declare readOnlyHint=true, indicating a safe read operation. The description adds value by specifying exactly what information is listed (supported quality, aspect-ratio, credit-cost). No contradictions.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two concise sentences: first states the purpose and specifics, second advises when to use. No redundant information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given no parameters or output schema, the description fully covers the tool's purpose and usage context. It is sufficient for an agent to understand and invoke correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
No parameters are defined, so the baseline is 4. The description is unaffected by parameter semantics.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool lists image and video models from ImgLume with details on quality, aspect ratio, and credit cost. It distinguishes from sibling tools that generate or search content.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly states 'Use when model choice, capability, or cost matters.' Provides clear context for when to use, though no exclusions or alternatives are mentioned.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search the ImgLume visual inspiration gallery. Use when the user wants ideas, references, styles, examples, or has not yet settled on a visual direction. Results include preview images and IDs.
| Name | Required | Description | Default |
|---|---|---|---|
| page | No | ||
| sort | No | popular | |
| limit | No | ||
| model | No | all | |
| query | No | Focused search words such as "luxury perfume poster". Leave empty to browse popular inspiration. | |
| category | No | Optional ImgLume category slug. | |
| mediaType | No | all |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint=true, so description doesn't need to emphasize safety. It adds that results include previews and IDs, but no details on pagination or rate limits. Provides some value beyond annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, no redundancy, front-loaded with key purpose. Every word adds value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Moderate complexity with 7 optional parameters and no output schema. Description covers basic purpose and usage scenario but not result format or parameter interactions. Sufficient but incomplete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is only 29% (query and category have descriptions). The description does not explain the other five parameters (page, sort, limit, model, mediaType) despite low coverage. Fails to compensate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool searches the ImgLume gallery for visual inspiration, specifying use cases like 'ideas, references, styles, examples'. It distinguishes from sibling tools like generate_image or list_models.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicitly says when to use: when user wants ideas or hasn't settled on direction. Implicitly suggests not to use for direct generation. Lacks explicit exclusions but context is sufficient.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0enhance_promptgenerate_imagegenerate_videoget_inspirationlist_modelssearch_galleryScored across 6 tools
Each tool has a clearly distinct purpose: searching, retrieving inspiration, enhancing prompts, listing models, and generating images or videos. No overlap in functionality.
All tool names follow a consistent verb_noun pattern using snake_case (e.g., search_gallery, generate_image), making it easy to predict tool behavior.
Six tools cover the essential workflow of visual inspiration and generation: search, get details, enhance prompt, list models, generate image/video. The count is well-scoped for this domain.
The tool set provides a complete lifecycle for visual generation: inspiration discovery, prompt refinement, model selection, and generation of images or videos. No obvious gaps.
Generate AI images and videos from any compatible MCP client.
Generate AI images, video, music, and sound effects, and upscale them, from any MCP client.
MCP server for Luma Dream Machine AI video generation
MCP server for Flux AI image generation
