Retro Diffusion Pixel Art MCP

Generate Pixel Art

create_inference

Generate images using the public /v1/inferences endpoint.

For the highest quality prefer RD Pro styles (rd_pro__*); they support reference_images for character/style consistency, and most go as small as 12x12 px (check list_available_styles for each style's limits) — a small target size is never a reason to switch to a cheaper model family. Style ids are opaque strings with no uniform format (some RD Fast styles appear as "default:rd_flux"); take them verbatim from the catalog and never infer capabilities from an id's prefix. For animation styles prefer start_inference_job + get_inference_job instead — animations are long-running, and a failed animation is worth one retry with identical parameters (failures auto-refund).

Field-tested workflow rules: N distinct items = N individually usable images (separate calls or num_images=N), never one sheet/grid image unless a sheet IS the deliverable. Variants of ONE image (seasons, day/night, palettes) = generate the base once, then derive each variant with the image_edit tool ("... keep the exact same composition") — independent generations of the "same" scene come out unrelated. Converting an existing image INTO pixel art is rd_pro__pixelate with input_image; reference_images-based generation re-imagines rather than converts. To animate an image you already have, use rd_advanced_animation__* with input_image (fixed-format rd_animation__* styles generate their own subject from the prompt instead).

Use input_image for the main source image, reference_images for extra per-inference guidance, and style_reference_images only on create_user_style/update_user_style. The response excludes raw base64 image payloads to keep MCP outputs compact.

Input Schema

TableJSON Schema

Name	Required	Description
`seed`	No	Seed for reproducible results; reuse the same seed to iterate on one composition.
`width`	Yes	Output width in pixels. Each style enforces its own limits; check list_available_styles or get_style_usage. Genre-native sizes per item: Minecraft 16; items/icons/props 32-64; character sprites 16-48 retro or 96-128 showcase; tiles 16-32; portraits 96-128; full scenes 256 (RD Pro's max; 16:9 scenes = 256x144 — pixel art integer-upscales losslessly).
`height`	Yes	Output height in pixels. Each style enforces its own limits; check list_available_styles or get_style_usage.
`prompt`	Yes	Describe the SUBJECT only, richly and concretely ('a squat round flask of glowing crimson liquid, cork stopper, bright highlight on the upper-left rim' beats 'a potion'). Never write 'pixel art' — the selected style handles all rendering. For standalone assets, state a flat background color that contrasts the subject (default 'on a plain white background') and pair with remove_bg=true; never write 'transparent background' (that is remove_bg's job), and never leave the background unstated (it drifts to drab dark gray). Scenes instead describe their real environment.
`tile_x`	No	Make the result tile seamlessly on the horizontal axis.
`tile_y`	No	Make the result tile seamlessly on the vertical axis.
`strength`	No	How strongly to change input_image, 0-1 (default 0.75). Lower values keep more of the original.
`remove_bg`	No	Remove the background for transparent output. Use true for standalone assets and pair it with a stated contrasting background in the prompt ('on a plain white background') — removal works best on flat contrasting backdrops. Animations inherit the start frame's transparency automatically.
`num_images`	No	How many images to generate in one batch; a batch produces varied takes of one prompt (the right way to get N distinct items as individually usable images — never pack N items into a single sheet/grid image unless a sheet IS the deliverable). Style-specific maximums apply.
`rd_api_key`	No	RetroDiffusion API key (rdpk-...) for this call only; overrides session or header auth.
`input_image`	No	Base64 PNG used as the main source image for edits, variations, tilesets, animations, or styles that require a starting frame. Raw base64 or a data URL. Send the NATIVE-resolution image: an upscaled display copy (e.g. a 96px sprite exported at 4x = 384px) exceeds style ranges and gets rejected — downscale to the true pixel grid first. For advanced animations the frame's dimensions must equal width/height, and sprites whose opaque pixels touch the canvas edge animate badly (pad onto a larger transparent canvas first, e.g. 48x48 content onto 64x64).
`extra_prompt`	No	Secondary prompt for styles that use one (e.g. the transition texture in advanced tilesets).
`prompt_style`	Yes	Style id from list_available_styles (e.g. 'rd_fast__default', 'rd_pro__isometric', or a custom 'user__...' style).
`input_palette`	No	Base64 image of a color palette; output colors are constrained to it.
`upload_outputs`	No	Host outputs and return URLs in output_urls (recommended for MCP clients) instead of only base64 payloads.
`frames_duration`	No	Animation frame count for animation styles: 4, 6, 8, 10, 12, or 16. Pick deliberately: 8 is the sweet spot for loops (walking, idle), 6 for a snappy single action, 10-12 for flowing ambient motion.
`timeout_seconds`	No	Read-timeout override in seconds for this call; increase for animations or large batches.
`reference_images`	No	Extra per-inference guidance images (base64), only for styles where supports_reference_images is true. Not for defining custom styles.
`extra_input_image`	No	Second base64 input image for styles that use one (e.g. the second texture in rd_tile__tileset_advanced).
`return_pre_palette`	No	Also return the render from before palette constraints were applied.
`return_spritesheet`	No	For animation styles: return a PNG sprite sheet instead of a GIF.
`return_non_bg_removed`	No	Also return the render from before background removal was applied.
`upscale_output_factor`	No	Integer upscale factor for the output image; 1 returns the native pixel size.
`bypass_prompt_expansion`	No	Skip the automatic LLM prompt enrichment and use the prompt verbatim.
`include_downloadable_data`	No	Include extra structured assets when available (e.g. tileset atlas JSON, animation frame data).

Output Schema

TableJSON Schema

Name	Required	Description	Default
No arguments

TDQS

A4.8/5.0

Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations (readOnlyHint=false, openWorldHint=true) only signal that this is a mutating external call; the description richly supplements them: failures auto-refund ('a failed animation is worth one retry'), the response excludes raw base64 to keep MCP outputs compact, and inputs have hard validation constraints (native-resolution only, sprite padding for animation edges). It also signals cost-relevant behavior (N items = N calls). No contradiction with annotations — it layers genuinely useful operation knowledge on top of them.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Long, but intentionally so — roughly 500 words across four organized movements (purpose → workflow → routing → parameter taxonomy). Every sentence carries hard operational weight (e.g., 'never infer capabilities from an id's prefix'); there's no filler. Slightly front-loading the field-tested rules might help an agent scanning quickly, but given 25 parameters and a family of sibling tools, the density is justified. Not a word is decoration.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Exceptional for a tool of this complexity: covers the sync-vs-async split (start_inference_job for long-running animations), error/retry semantics (auto-refund), output behavior (base64 vs output_urls, no raw base64 in responses), and edge cases (sprites touching canvas edges animate badly, upscaled copies get rejected). With an output schema present, nothing about the return contract is missing. An agent has everything needed to invoke this correctly and — just as important — to know when NOT to invoke it.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema covers 100% of 25 parameters, so the baseline is 3 — and indeed the schema's own per-parameter descriptions are exceptional. The description earns its keep by adding CROSS-parameter semantics the schema cannot: the paragraph dictating that input_image is for the main source, reference_images for per-inference guidance, and style_reference_images only on create_user_style/update_user_style prevents a class of parameter-misuse errors no single-param schema could catch. Only a hair below 5 because much of the parameter-heavy lifting (style sizes, frame counts) lives in the schema where credit was already given.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Opens with a specific verb+resource ('Generate images using the public /v1/inferences endpoint') and immediately scopes the tool's identity against siblings like start_inference_job and run_edit_tool. The description differentiates it clearly from the animation path ('prefer start_inference_job + get_inference_job instead') and from image_edit, making it unambiguous when this tool is or isn't the right choice — far beyond a tautological restatement of the name.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Superb when-to-use and when-not-to-use guidance: animations route explicitly to start_inference_job/get_inference_job with the reasoning ('long-running'); pixelating an existing image routes to rd_pro__pixelate with input_image; transforming an existing image routes to rd_advanced_animation__*; variants of one image route to image_edit. The 'Field-tested workflow rules' paragraph names concrete alternative tools and the conditions that select them — the gold standard for usage disambiguation.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

All Retro Diffusion Pixel Art MCP tools Inspect

Try in Browser

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

TDQS

A4/5.0

Disambiguation4/5

Most tools have clearly distinct purposes (e.g., create_inference vs start_inference_job vs get_inference_result). However, the difference between create_user_style/update_user_style and per-inference references could still cause confusion, and list_available_models/list_available_styles overlap slightly.

Naming Consistency5/5

All tool names consistently follow a verb_noun pattern (e.g., create_inference, get_balance, list_edit_tools, delete_user_style). No mixing of camelCase or other styles, making the surface highly predictable.

Tool Count5/5

With 20 tools covering authentication, inference (sync/async), styles, editing, cost estimation, and status, the count is appropriate for a pixel art generation API. Each tool addresses a distinct need without bloat.

Completeness4/5

The tool set covers the full lifecycle: auth, cost estimation, synchronous/async generation, style management, editing, and result retrieval. A minor gap is the lack of a tool to list or manage user styles (e.g., get_user_styles), but this is non-critical for core workflows.

Resources

Need Help?