Integrations

Claude (MCP)

SteadyShot exposes its full API as a set of MCP tools that Claude can call by natural language — generate characters, scenes, storybooks, and variations without leaving the chat. Two ways to connect: claude.ai custom connector (browser, one-click) and Claude Desktop (config file).

Brand-kit tools below (list_brand_kits, get_brand_kit, generate_with_brand) are in private beta and require beta access on your account. Request access →

Option A · claude.ai custom connector (recommended)

Works in the browser and on the claude.ai mobile app. No CLI, no config files — paste a URL + your API key once.

Grab an API key at steadyshot.ai/api-keys.
In claude.ai, open Settings → Connectors → Add custom connector.
Use these values:

text

Server URL:  https://steadyshot.ai/api/mcp
Auth header: Authorization: Bearer ss_live_xxxxxxxxxxxxxxxx

Save. The connector appears in your tools tray in any new chat. Ask “list my SteadyShot characters” to confirm it's wired up.

The connector inherits your API key’s scope. To restrict it to a single workspace, send the X-Steadyshot-Workspace header from the connector config; otherwise it defaults to your personal workspace.

Option B · Claude Desktop (stdio)

For Claude Desktop, the local stdio server is still the simplest path — fewer round trips, no internet required for tool dispatch (the tool handlers still hit the SteadyShot API).

Grab an API key at steadyshot.ai/api-keys.
Open Claude Desktop’s settings → Developer → Edit Config.
Add the SteadyShot MCP block to claude_desktop_config.json:

json

{
  "mcpServers": {
    "steadyshot": {
      "command": "npx",
      "args": ["-y", "steadyshot-mcp"],
      "env": {
        "STEADYSHOT_API_KEY": "ss_live_xxxxxxxxxxxxxxxx"
      }
    }
  }
}

Restart Claude Desktop. The 🔌 icon at the bottom of the chat should show steadyshot connected with 12 tools available.

Running from source instead? Point command at node and args at the absolute path of mcp-server.mjs in your local clone.

What you can ask

“List my characters” → list_characters
“Generate a scene of Luna reading by a campfire” → generate_scene
“Make 4 frowning variations of that scene” → vary_scene
“Generate a 5-page storybook of Luna meeting Milo at the seaside” → generate_storybook
“Combine Luna with the Acme brand kit for a marketing shot” → generate_scene with both ids
“Generate reference sheets for character X” → generate_references

All tools

Tool	What it does
`list_characters`	List all your characters.
`get_character`	Get one character + references.
`generate_scene`	Generate a single-character (and/or brand) scene.
`vary_scene`	Create 1-4 single-dimension variations of an existing scene.
`generate_multi_scene`	Scene with 2-4 characters together.
`generate_storybook`	Multi-page storybook with optional PDF export.
`generate_references`	Generate reference sheet views for a character.
`list_brand_kits`	List your brand kits. (beta)
`get_brand_kit`	Get one brand kit + assets. (beta)
`generate_with_brand`	Brand-only scene generation (no character). (beta)
`delete_character`	Permanently delete a character.

Credits + auth

Every tool call bills against the credits of the user whose API key is in your MCP config. Same plan limits, same per-action costs as the web UI — Claude doesn’t get any special billing.

Sample conversations

These are real prompts that exercise multi-turn workflows. Claude figures out which tool to call from context and threads image_ids automatically.

Conversation 1 — character → scene → variation

text

You: "Show me my characters"
↳ list_characters → ["Mira (locked)", "Beanie (locked)", "Kai (draft)"]

You: "Generate Mira sitting by a moonlit pond"
↳ generate_scene(character_id=mira-id, prompt="Mira sitting by a moonlit pond")
→ returns image_id + url, Claude shows the image

You: "Make 4 versions with different expressions"
↳ vary_scene(image_id=<prev>, vary="expression", target="varied", count=4)
→ 4 thumbnails inline

You: "I like #3 — now make her at golden hour"
↳ vary_scene(image_id=<#3 image_id>, vary="lighting", target="golden hour", count=1)

Conversation 2 — full storybook from scratch

text

You: "Generate a 4-page bedtime story for Mira:
  1. She wakes up in a forest clearing
  2. She finds a glowing mushroom
  3. She touches it and it pulses with light
  4. She runs home through the trees, smiling
  Style: studio ghibli watercolor. Export to PDF."

↳ generate_storybook(
    character_id=mira-id,
    pages=[
      "She wakes up in a forest clearing",
      "She finds a glowing mushroom",
      "She touches it and it pulses with light",
      "She runs home through the trees, smiling"
    ],
    art_style="studio ghibli watercolor",
    export_pdf=true
  )
→ returns 4 page images + a pdf_url

Conversation 3 — brand kit + character together

text

You: "Generate Beanie holding an Acme Coffee cup at a sunlit café"
↳ list_brand_kits → ["Acme Coffee"]
↳ generate_scene(
    character_id=beanie-id,
    brand_kit_id=acme-id,
    prompt="Beanie holding an Acme Coffee cup at a sunlit café"
  )
→ image with the exact Acme logo on the cup, brand palette obeyed

Build your own MCP-compatible client

The SteadyShot MCP server is a standard Model Context Protocol stdio server. Any MCP-compatible client can connect — not just Claude Desktop.

Server source: mcp-server.mjs in the SteadyShot repo. Single file, ~370 lines, no build step.
Tools registered: see the table above. Each tool is declared in the tools array and routed via CallToolRequestSchema.
Transport: stdio (newline-delimited JSON-RPC). Wrap with your own MCP client SDK to call from a custom agent.

← Previous

REST API

Figma plugin