Integration · MCP

Your editor can read the library directly

MCP is the protocol an AI editor uses to call outside tools. Add this server once and your agent stops guessing at component names: it searches the catalog, reads the scores, and pulls the actual source — the same records these pages render, read at request time.

The endpoint

https://www.56juqingba.com/api/mcp

Transport: Streamable HTTP · JSON-RPC 2.0
Protocol revision: 2025-06-18
Auth: None — it is a public read

Install

Paste the block below into the file your client reads, then restart it. No API key, no account, nothing to sign.

Cursor

.cursor/mcp.json in your project, or ~/.cursor/mcp.json globally

{
  "mcpServers": {
    "motif": {
        "type": "http",
        "url": "https://www.56juqingba.com/api/mcp"
      }
  }
}

Claude Desktop

claude_desktop_config.json — Settings → Developer → Edit Config

{
  "mcpServers": {
    "motif": {
        "type": "http",
        "url": "https://www.56juqingba.com/api/mcp"
      }
  }
}

Windsurf

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "motif": {
        "type": "http",
        "url": "https://www.56juqingba.com/api/mcp"
      }
  }
}

5 tools

This list is generated from the definitions the endpoint serves, so the page cannot advertise a tool that does not answer.

search_assetsSearch the component library
Search 362 original animated components by keyword. Matches the slug, title, description, tags and recorded aliases. Returns a compact list; call get_asset for the full record.
Returns: slug, title, kind, stacks, bundle size, dependencies, both scores, licence and the page URL.
q *kindstacklimit
get_assetGet one component
The full record for one component: scores, dependencies, bundle size, props and the URLs of its page, live demo and JSON API.
Returns: every recorded field for that asset, including the prop list with defaults.
slug *
get_snippetGet a component's source
The verbatim source of one component's demo, its paste-ready call site and, where one is recorded, a standalone stylesheet. This is the code the detail page shows.
Returns: source, usage and css as text, plus the module each was extracted from.
slug *part
search_promptsSearch the AI prompt library
Search 184 website prompts. Every prompt carries recorded runs against named models, including the runs that failed; get_prompt returns them.
Returns: slug, title, industry, vibe, stacks, blocks, average fidelity, best model and run count.
q *industrylimit
get_promptGet one prompt with its run log
The full prompt text plus every recorded run: model, date, fidelity score, whether the build errored, and the note written at the time.
Returns: the verbatim prompt body and the complete run list, failures kept.
slug *

* required

What the server will and will not do

Read-only. There is no tool that writes, because nothing on this site should change on a visitor's behalf.
Stateless: no session is created and nothing about a previous call is remembered.
The endpoint answers JSON-RPC over POST and opens no event stream, so a client that insists on SSE gets a 405 and can still call every tool.
Every number it returns is the number the site's own pages print — the same records, read at request time.

Try it from a terminal

No client needed — the endpoint answers plain JSON-RPC, so one curl is enough to see whether it is up.

curl -s https://www.56juqingba.com/api/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

← All integrations REST API docs Browse the library