@floats/mcp

MCP server for Floats — give any AI app read/write access to your canvases: cards, text boxes, quotes, code blocks, section breaks, page breaks, margin notes on any float (images included), workspace search, and spaced-repetition flashcards.

Get an API key in your Floats dashboard → API keys.

Remote first (no install)

Most modern clients (claude.ai, Claude Code, ChatGPT, Cursor) can connect straight to the hosted endpoint:

https://thefloats.app/api/mcp

with Authorization: Bearer flt_…. Claude Code:

claude mcp add --transport http floats https://thefloats.app/api/mcp \
  --header "Authorization: Bearer flt_YOUR_KEY"

Clients without header support can append ?key=flt_YOUR_KEY to the URL.

Local stdio (this package)

For stdio-only clients (Claude Desktop, Gemini CLI, …):

{
  "mcpServers": {
    "floats": {
      "command": "npx",
      "args": ["-y", "@floats/mcp"],
      "env": { "FLOATS_API_KEY": "flt_YOUR_KEY" }
    }
  }
}

Tools

list_canvases · read_canvas · create_canvas · add_floats (cards / images / text boxes / quotes / links / sections / code / margin notes + flow arrows) · add_page · update_float · delete_float · search_workspace · due_flashcards · create_flashcards · grade_flashcard

Everything you add appears on the canvas live, in the app's own shapes.

Building code-side agents instead? Use @floats/sdk — a typed client plus Vercel AI SDK tools that run on any model.

MIT © Floats