Canvora

Install as a skill

npx skills add canvora/canvora-mcp

Canvora — visuals for AI agents

[npm] [Canvora MCP server] [License: MIT]

Give your agent a design department. Turn any idea, URL, doc, or PDF into finished, on-brand visuals: 100+ formats (posts, carousels, decks, PDFs, ads, infographics, landing-page kits), generated natively in 150+ languages — including RTL and CJK. Brand kits keep every output in the client's colors, fonts, logo, and tone.

canvora generate --idea "5 tips for remote work" \  --format linkedin_carousel --slides linkedin_carousel=5 \  --brand $BRAND_ID --wait --json# → 5 finished branded slides on CDN URLs

Your agent writes the brief. Canvora does the design. The outputs are stable CDN URLs ready to post, embed, or hand to a scheduler (works great upstream of Postiz or any posting tool: generate → schedule, end to end agentic).

This repo is the agent-facing toolkit for Canvora:

• Agent skills — skills/: 12 task-specific skills, one npx skills add installs all of them into Claude Code, Cursor, Codex, OpenCode, OpenClaw and 60+ agents.

  Skill	For
  canvora	core: setup, generate loop, credits, refinement
  instagram-carousel	Instagram carousels / swipe posts
  linkedin-carousel	LinkedIn document carousels + safe zones
  pitch-deck	pitch/sales/training decks, document → slides
  pdf-document	multi-page PDFs: brochures, reports, one-pagers
  landing-page-visuals	a whole page's visual kit from its URL
  ad-campaign	paid social + display ad sets and variants
  blog-promo-kit	header, OG image and promo set from one post
  data-infographic	process flows, comparisons, timelines, stats
  multilingual-campaign	one campaign → many languages
  product-photo-visuals	designs built around the user's own photos
  visual-refresh	edit, vary, localize and revert what exists

• Remote MCP server — for Claude, ChatGPT, Cursor, and any MCP client. Nothing to install.

• canvora CLI — for shell-native agents, cron jobs, and CI. Zero dependencies.

The Canvora service itself is hosted; this repo contains client tooling, skills, and docs.

Option 1 — Skills + CLI (Claude Code, Cursor, Codex, OpenClaw, shell-native agents)

npx skills add canvora/canvora-mcp     # installs the skills (they set up the CLI on first use)# or just the CLI:npm install -g @canvora/cli
export CANVORA_API_KEY=vd_...   # create at canvora.ai → Integrations → API Keys (all plans, incl. Free)
# one atomic command: generates, polls, exits with the output URLscanvora generate \  --idea "5 tips for remote work" \  --format linkedin_carousel --slides linkedin_carousel=5 \  --wait --json

More runnable recipes: examples/

Commands

Every command takes --json for machine-readable stdout; progress and errors go to stderr, so piping stays clean. Exit codes: 0 success, 1 failure, 2 usage error.

Generate flags

--idea "concept"          short concept - Canvora develops it into content--input "text"            your full content text, used as the source material--url <url>               extract content from a webpage--file-url <pdf-url>      extract content from a PDF--format <id>[,<id>...]   output formats (see: canvora formats)--slides <format>=<n>     slide count for carousel/deck formats (1-10, default 5); match it to the item count your idea promises--brand <uuid>            apply a brand kit--style <name>            modern | minimal | bold | elegant | playful | corporate | creative | dark--resolution 2K|4K        4K on paid plans--language <code-or-name> visual text language ("es", "Spanish"); default: matches your input's language--exact                   use --input text verbatim, no rewording (quotes, taglines)--logo                    place the brand kit's logo on the visual (needs --brand)--cta                     end a carousel/deck on a call-to-action slide--no-text                 pure visual, no typography (backgrounds, hero imagery)--freeform                ignore format-specific art direction, use dimensions only--landing-page            with --url: read the page's sections and produce a matching                          landingPage-* visual kit (Starter plan or higher)--project <uuid>          group outputs into a project/campaign--workspace <uuid>        bill a team workspace pool instead of your personal credits--wait                    poll until done (--timeout 600, --poll 5 seconds)--json                    machine-readable output

Option 2 — MCP (Claude, ChatGPT, any MCP client)

One remote server, OAuth sign-in, 16 tools your agent discovers on connect:

Step-by-step: canvora.ai/help/integrations/mcp-setup

stdio-only clients: mcp/ is a small stdio server that forwards every call to the remote server with your API key (same 16 tools, nothing generated locally):

{  "mcpServers": {    "canvora": {      "command": "node",      "args": ["/path/to/canvora-mcp/mcp/server.js"],      "env": { "CANVORA_API_KEY": "vd_..." }    }  }}

Run npm install in mcp/ first, or use the Dockerfile: docker build -t canvora-mcp . && docker run -i -e CANVORA_API_KEY=vd_... canvora-mcp.

Option 3 — Raw REST

curl -X POST https://api.canvora.ai/api/generations \  -H "X-API-Key: vd_YOUR_KEY" \  -H "Content-Type: application/json" \  -d '{    "inputType": "text",    "inputContent": "5 tips for remote work",    "outputFormats": ["linkedin_carousel"],    "carouselCounts": {"linkedin_carousel": 5}  }'

Full API guide: canvora.ai/help/integrations/api-keys

What agents build with this

• Client content pipelines — brief in, branded multi-format campaign out, scheduled by your posting tool
• Multilingual campaigns — generate once, canvora localize into 10 markets; layout survives, text goes native (skill)
• Save-optimized carousels — hook-first structure and per-slide art direction encoded as a skill (skill)
• Docs → decks — point --file-url at a PDF report, get presentation slides
• Ad sets — ad_square,ad_landscape,ad_portrait from one product URL, all placements covered

Built to be called by software

• Predictable costs — flat credit pricing (10/visual, 15/slide); canvora credits before big jobs; failed generations auto-refund.
• Human in the loop — everything your agent creates also lands in your Canvora dashboard, ready for review, editing, and export before it goes anywhere.
• Scoped, revocable keys — hashed at rest, per-key usage tracking, one-click revoke.
• Outputs on a CDN — stable URLs your agent can post, embed, or hand to a scheduler.
• Works on every plan — API and MCP access included on Free (free credits, no card).

Links

• canvora.ai/for-agents — the agent overview
• MCP setup guide · API guide
• Pricing

MIT licensed (CLI, skills, and docs in this repo).