Magnific MCP Server

User plan + credits. Check before paid generations. Returns plan.{tier,productName,isUnlimitedMode}, credits.{available,totalPlan,spent,hasExtraCredits}. No history.

AI music generation. Google Lyria or ElevenLabs. Pass `prompt`, optional `model`/`durationSeconds`/`instrumental`. Lyria/Lyria-3 produce fixed 30s; Lyria-3-Pro accepts 30–180s; ElevenLabs accepts 10–300s.

TTS voiceover. Single (voiceId) or 2-speaker (speakers). ElevenLabs/Google. No voiceId? Call `audio_voices_show` first (or `audio_voices_list` for agent-only lookup).

TTS voice catalog as raw TOON text, no UI. Lean entries without preview URLs. Optional `search` filters by name, description, gender, or language.

Add a comment (or reply) to a user-owned creation.

Step 2 after PUT from `creations_request_upload`: convert temp path(s) to creations. Images <=25MB, videos <=200MB.

Single creation complete data — metadata + URLs (`url` full-res, `previewUrl` ~1024px, `thumbnailUrl` ~400px).

Toggle heart/favorite reaction on a user-owned creation.

Filterable gallery widget (image/video/audio). Use instead of `creations_search` when user wants to browse their library.

Move creations into `targetFolderReference`. Identifiers from `images_generate`/`creations_search`. Foreign identifier → abort. Target folder must belong to user.

Step 1 for local/user files: get presigned PUT URL(s). Upload bytes outside MCP, then call `creations_finalize_upload`.

Search user creations as raw TOON text. Use `creations_get` for one id. `reference` required with from=folder/project-root. Defaults: history, page 1.

Import a file the user attached in the host (e.g. a ChatGPT upload) as a creation, then use the returned identifier in image tools. Use this instead of asking for a URL when the user already uploaded a local file. For a public URL use `creations_upload_image`.

Upload public image URL in one step. For local/user files use `creations_request_upload` then `creations_finalize_upload`.

Long-poll 1..8 creations until terminal or `timeoutSeconds` (default 25, max 25). In-progress entries get `poll_after_seconds` for the next call.

Extract editable layers from a flat image into a Pikaso design.

Create resized design pages from an image. Use directly when ratios/social formats are known; otherwise show `design_auto_resize_show`.

Flow input/output spec as raw TOON text. Each input includes a normalized `kind`, `howToProvide`, and resolved `options`/`presetOptions` where applicable. Call before `flows_run`.

Flows catalog as raw TOON text, no UI. Lean entries without preview URLs. For an inline picker use flows_show; for a single flow input/output spec use flows_get.

Run flow. `flows_get` first for input specs. `inputs` is `{inputId: value}` (creation inputs from `creations_list`, voices from `audio_voices_show`). Returns `workflowRunIdentifier` → `flows_wait` → `creations_show`.

Long-poll a `flows_run`. Terminal: one entry per output creation with real `identifier` for `creations_show`. In-progress: `poll_after_seconds`.

Create folder/project. No `parentReference`: workspace root. `type` defaults: project at root, folder when nested.

Move a folder and its files to trash.

Fetch folder metadata by `reference`. Use before `creations_search` (`from=folder`). Foreign refs error.

List folders/projects as raw TOON text. No params=top level; `onlyProjects`=projects; `parentReference`=children. Refs work with creations/folder tools.

Rename a folder by `reference`.

Reframe camera around subject. Defaults: rotate=45, vertical=0, closeup=5.

Center-crop to `aspectRatio`. Pixels: `images_resize`. AI upscale: `images_upscale`. Ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9.

Generate/edit images. No refs=TTI. references[] (max 12; see field docs). Defaults: mode=auto, ratio=1:1, count<=8. resolution/quality per images_models_list. UI clients: after generating, MUST call `creations_show` with all identifiers; never stop at links.

Text-to-SVG (Recraft v4 Pro Vector). Raster: `images_generate`. Trace existing raster: `images_to_svg`. Default aspectRatio=1:1.

TTI model catalog as lean TOON text. `search` filters by name, slug, description, or tags.

AI relight. 1-4 lights with azimuth/elevation enums, type neutral|gel, intensity 1-10. Gel uses hex `color`.

Cut-out subject on transparent PNG. Color BG: `images_color_background`. AI scene: `images_generate` with reference + prompt "replace the background with X".

Resize to exact pixels (no AI). For real detail upscaling: `images_upscale` first. Aspect: `images_crop`. Dimensions snap to multiples of 8; center-crop on aspect mismatch.

AI skin enhancer. faithful preserves identity; creative reinterprets; flexible uses presets. Defaults: faithful, sharpen=0, smartGrain=0.

Trace raster creation into SVG. New vector from prompt: `images_generate_svg`.

AI upscale 2x/4x (Magnific, premium). Exact pixels: `images_resize`. Aspect: `images_crop`. Default scale=2.

Generate image-variation grid. Modes: angles|demographics|expressions|age|storyboard|custom. Defaults: angles, 3x3, 1:1, 4k.

Create a reusable library asset (character/product/location) from 1-6 images. Use the returned numeric `id` in generation refs (same id Spaces uses). No LoRA training.

Lean library catalog (characters, styles, elements, locations) for agent reasoning. Picker assets use library://.

Image-to-3D GLB. Upload external images first. Animate with `models3d_rig`. Models: tripo-p1 default, tripo-v31, trellis-2.

English markdown status report for a project/folder. Last 7 days default, max 30.

Create empty Space. Share `space.webUrl`; only call `spaces_show` when inline preview is requested.

External MCP only. Start headless Space edit. Poll `spaces_edit_status` until allTerminal, then verify with `spaces_state`.

Wait for `spaces_edit` like `creations_wait`: long-polls up to `timeoutSeconds` (default 25, max 25). Snapshot-only when 0. Terminal → `spaces_state`.

Read-only. Like `spaces_state` but scoped to `nodeIds` + their connections. Cheaper than full board. Needs view perm.

Lean spaces catalog as raw TOON text. Share each `webUrl`; call `spaces_show` only for inline canvas preview. Do not use `creations_search`.

Run Space workflow from `startNodeId`. Modes: singular, connected default, downstream. Poll with `spaces_run_status`.

Poll `spaces_run`. `timeoutSeconds` long-polls up to 25s. In-progress returns `poll_after_seconds`; terminal call `creations_show`.

Read-only board context, raw TOON. Default: current page + pageIndex (other pages, counts). pageId: another page on demand. scope=all: whole board (costly). selectedElementIds: selected nodes. Needs view perm.

Signed download URL for a stock item. Pass folder_reference to also save as a creation (rate-limited). Bulk support: accepts ids for batched execution.

Full details for a single Freepik stock item: preview URL, dimensions, formats, author. Bulk support: accepts ids for batched execution.

Search Freepik stock catalog by keyword. Returns items with IDs, titles, types, preview URLs. Use stock_get for full details, stock_download for file URL.

Convert Freepik stock item → Pikaso creation. Use after user selects from stock_show. Bulk support: accepts ids for batched execution.

Concatenate completed video creations into one MP4. Assembles multi-clip plans from `video_plan`. Audio from each clip; no external audio bed. Returns queued creation.

Generate video. External clients: call `video_plan` first to draft the brief and resolve model choice — skip only if the user explicitly says "just generate" or "one-shot". Pick a model with `slug` (copy it verbatim from `video_models_list`); omit `slug` for auto-select. Image refs: use an asset URL or a creation `identifier`; never `webUrl`. Limits: `video_models_list`. On UI-capable clients, after generating you MUST call `creations_show` once with all returned identifiers to render results; never stop at links.

Video-gen model catalog as lean TOON text. `search` filters by name, slug, description, or tags. Run `video_plan` first — plan resolves slug to validate here.

Call FIRST for any video request. Returns markdown plan: brief, open questions, characters to prepare, recommended model slug, prompt draft. Videos >15s: plan splits into clips + `video_concatenate` instructions.

Make a character speak. Image+audio → talking head (Veed Fabric 1.0). Video+audio → lip sync (Lipsync 2.0). Omit `mode` for auto. Pass asset URLs or a creation `identifier` (from `audio_tts`, `images_generate`, `video_generate`). Then call `creations_show`.

Upscale video via Topaz or Magnific. Pass `creationIdentifier` or Freepik `videoUrl`. Params: `video_upscale_models_list`. Then call `creations_show`.

Video upscale mode catalog as raw TOON text (topaz, magnific, magnific_precision) with required/optional params, ranges and enums. Read before calling `video_upscale`.

MCP.AI for IDE agents (Cursor, etc.): log in in the browser, copy the access token. Best: add it to this server's config as a header `Authorization: Bearer <token>` for a permanent, non-expiring connection. Or paste it here for a session-only login: call with { token: "<jwt>" } after the user pastes, or with no args to get the link.

Returns connection status and URLs. When all providers are connected, returns authenticated:true and empty pending[]. When credentials are missing, returns connect_url for the toolkit and per-install URLs.

The official mcp.ai marketplace — the in-platform catalog of every MCP/tool, AND the way to run them. Covers capability requests like "find an MCP that does X", "consulta um CPF", "is there a tool for Y". Core flow: action=search discovers MCPs by intent → describe returns one MCP's full profile (every tool with its id + params, pricing, auth) so you pick the right tool_id → invoke RUNS that tool. KEY: invoke works even when the MCP is NOT installed — it runs the tool pontualmente (one-off), without adding the MCP to the toolkit and without bloating the tool list. If the MCP needs a credential/login, invoke returns a connect link; if it is paid and the wallet is empty, invoke returns a checkout/top-up link (the user opens it, then you retry). Use install only to make an MCP PERMANENT in the active toolkit (its tools then show up natively in future sessions); prefer invoke for a single/occasional use. list_tools lists what is callable right now. subscribe/cancel handle per-MCP billing; report_bug sends feedback; request_mcp asks us to build a NEW MCP when nothing fits. Search/describe flag installed_in_toolkit vs installed_in_workspace. Writes (install/uninstall/subscribe/cancel and the one-off install behind invoke) require workspace owner/admin. It also carries the mcp.ai PROMPT LIBRARY, which is about ready-made prompt TEXT rather than MCPs: search_prompts finds one, get_prompt returns its full text with {{variables}} filled, and publish_prompt saves a prompt and returns a shareable mcp.ai/p/<slug> link that opens without login.

Report a bug, missing feature, or send feedback. Include the conversation array with recent messages for reproduction.

Show the current MCP platform and adapter versions.

Returns the current toolkit state: installed MCPs, their connection status, the accounts connected to each one, and how many catalog tools each exposes.