StockCake

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Search millions of CC0 public-domain AI images: previews, dimensions, provenance, license.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "StockCake": {
      "url": "https://stockcake.com/api/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (13)

• search_images read-only

  Natural-language search over StockCake's catalog. total_results is the total number of matches across the whole catalog, not just this page. refinements lists exact mood/style values for this query with counts (how many current matches carry each); both are exact filters, so a small count means few results. Start with a short query; when total_results is small or off-target, retry with fewer keywords or different phrasing.

  Parameters:

  • query (required) — string — Short subject, 2–5 words (e.g. 'red vintage rocket'). Long sentences hurt recall: start short, add one qualifier at a time; if results are thin, drop keywords or rephrase.
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • mood — string — Exact facet value — pick one from refinements.mood of a previous search_images result (check refinements.counts.mood first: it is a hard filter, not a boost)
  • page — integer — Page number, 1-5
  • type — string
  • color — string — Dominant color, e.g. a hex value like #ff0000
  • limit — integer — Results to return, 1-12 (default 12). Ask for fewer to save tokens.
  • style — string — Exact facet value — pick one from refinements.style of a previous search_images result (check refinements.counts.style first: it is a hard filter, not a boost)
  • locale — string — Site locale for the user (default en). Titles and page links are localized; descriptions are English.
  • orientation — string
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• find_similar_images read-only

  Find images visually similar to a given StockCake image id (from search_images results). total_results is the number of similar results returned on this page, not a catalog-wide count.

  Parameters:

  • id (required) — integer — Numeric image id from a previous result
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • page — integer — Page number, 1-5
  • limit — integer — Results to return, 1-12 (default 12). Ask for fewer to save tokens.
  • locale — string — Site locale for the user (default en). Titles and page links are localized; descriptions are English.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• get_image read-only

  Fetch one image with full metadata: an inline preview the model can see (unless include_preview is false), real pixel dimensions, AI-generation provenance (prompt, dates, style), license details, and attribution snippets. Also shows the signed-in user's own edited images (ids ≥ 10000000000, as returned by generate_image, resize_image and upscale_image).

  Parameters:

  • id (required) — Numeric image id (from search results), one of your own edited image ids (from a generate / resize / upscale result; needs the signed-in account), or the image's CDN uuid
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • locale — string — Site locale for the user (default en). Titles and page links are localized; descriptions are English.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.
  • include_preview — boolean — Default true. Set false to skip the inline preview image and save tokens.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• get_license_info read-only

  License and usage terms for all StockCake images (site-wide CC0 / public domain).

  Parameters:

  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• get_account read-only

  The signed-in user's plan, credit balance (free / subscription / purchased), next resets, download quality and output formats. Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• list_models read-only

  Image and video models available to generate_image / create_video with the credit price of every tier. Free-credit models are flagged; video is priced per second. Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • kind — string — Return only one catalog
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• generate_image

  Generate a new image from a prompt with one of StockCake's AI models (list_models). Charges credits; free credits cover the free models. Waits up to 45 s — a pending result carries a job_id for get_job. Optional reference images guide the composition (plan-capped). Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • prompt (required) — string
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • tier — string — Tier key from list_models (e.g. 1k, 2k, standard). Default: the model's default tier.
  • model — string — Model id from list_models. Default: the site default for subscribers, the free model otherwise.
  • aspect_ratio — string
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.
  • reference_images — array — Stock image ids or your own edited image ids used as references

  Hints: read-only no · destructive no · idempotent no · open-world yes

• upscale_image

  Upscale a StockCake image or one of your edited images 2x, 4x or 8x (lossless PNG output). Plan quality caps apply; charges credits otherwise. Waits up to 45 s, then returns a job_id for get_job. Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • image (required) — A StockCake image id (from search_images), one of your own edited image ids (from a previous result), or a job_id. Never a URL.
  • scale (required)
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only no · destructive no · idempotent no · open-world yes

• resize_image

  Change an image's aspect ratio by extending the canvas with AI (GPT Image 2.5) — the subject is kept, new area is generated. Charges credits. Output is the model's native size — about 1.5 megapixels, up to 1536 px on the long edge — whatever the source size; the result reports exact dimensions. For print or larger web use, run upscale_image on the result. Waits up to 45 s, then returns a job_id for get_job. Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • image (required) — A StockCake image id (from search_images), one of your own edited image ids (from a previous result), or a job_id. Never a URL.
  • aspect_ratio (required) — string — Target ratio
  • prompt (required) — string — What should fill the extended canvas, e.g. 'extend the sandy beach and sky; keep the subject unchanged, do not crop or stretch'
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only no · destructive no · idempotent no · open-world yes

• create_video

  Generate a short video from a prompt and/or a start image with one of StockCake's video models (list_models). Priced per second of output; needs a plan. Video usually takes longer than 45 s: expect a pending result and poll get_job. Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • tier — string — Tier key from list_models (e.g. 720p, 1080p, 720p-audio)
  • model — string — Video model id from list_models (default: H3 Max)
  • prompt — string — Required unless start_image is given
  • start_image — Animate from this image (stock id or your edited image id)
  • aspect_ratio — string — Text-to-video only; a start image sets the ratio
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.
  • negative_prompt — string
  • duration_seconds — integer — Clip length; must be one of the model's durations

  Hints: read-only no · destructive no · idempotent no · open-world yes

• download_image read-only

  Full-resolution download link for a StockCake image (CC0) or one of your edited images. Unmetered — no credits. format "png" returns the lossless sibling when one exists (falls back to jpg). Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • id (required) — Stock image id or your own edited image id
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • format — string — Default jpg (the original)
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• get_job read-only

  Status and result of a generate_image / upscale_image / resize_image / create_video job you started (pending, completed with URLs, or error with refund state). Requires a free StockCake account (the host will prompt to sign in).

  Parameters:

  • job_id (required) — string
  • context (required) — string — In one sentence, what is the user trying to make or find?
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

• send_feedback read-only

  Send feedback about this server to its developers. Most important: report a missing capability whenever no available tool fits your task, even if you can work around it (feedback_type "missing_capability"). Also welcome: a tool that failed or confused you, an unhelpful error, or something that worked well. This records the feedback; it does not add or change tools. Do not include user PII or sensitive content.

  Parameters:

  • feedback_type (required) — string — What kind of feedback this is. Use 'missing_capability' when the tool you needed does not exist in the tool list - nothing failed, the capability is absent (this is the most valuable report; send it even if you found a workaround). Use 'issue' when an existing tool behaved badly: it failed, returned a confusing error, or its description or schema misled you. Use 'praise' when something worked notably well. Use 'other' for anything else.
  • summary (required) — string — One self-contained sentence. For 'missing_capability': the capability you needed, e.g. 'No tool to delete multiple cohorts in one call.' For 'issue': the tool and the problem, e.g. 'query-trends rejects relative date ranges with an unclear error.'
  • llm_model (required) — string — The exact model identifier you (the assistant) are running as, taken from your system prompt or environment (e.g. "claude-opus-4-8", "gpt-5.2"). Used for analytics only. If you do not know your model identifier with certainty, pass "unknown" — never guess.
  • details — string — Optional longer context: what you tried, exact parameter values, the error text you saw, and any workaround you used. Omit when the summary says it all.
  • sentiment — string — Optional: how the experience felt overall.
  • tool_name — string — Optional: the existing tool this feedback is about (for 'issue' or 'praise'). Leave empty for 'missing_capability' - the point is that no tool fits.
  • task_completed — boolean — Optional: whether you still completed the user's task despite the problem.
  • conversation_id — string — Pass the exact conversation_id from the server's previous response, unchanged. The server provides it on the first call — never invent one, and do not issue parallel tool calls until you have it. Keep passing the same conversation_id for the rest of the conversation, including after later user messages or on a different task; do not reset it when the user starts a new request.
  • friction_points — string — Optional: the specific moments that slowed you down, as short bullet-like sentences, quoting exact tool names, parameters, or error text.
  • suggested_improvement — string — Optional: the concrete change that would have helped, e.g. the tool to add, the description to reword, or the error message to improve.

  Hints: read-only yes · destructive no · idempotent yes · open-world yes

Publisher website

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.