Skip to main content

Implement & Review with AI (MCP)

Fast Simon MCP — build and review the SDK with your AI assistant​

The Fast Simon MCP server gives AI coding assistants authoritative, always-current knowledge of the Fast Simon JavaScript SDK — real method signatures pulled straight from the shipped type declarations, the documentation, and the reporting events — so your assistant can implement and review your Fast Simon integration accurately, right inside your own project.

It works with Claude Code and any other MCP-capable assistant (Cursor, Windsurf, Cline, …).

Add it once​

Claude Code:

claude mcp add --transport http fastsimon https://mcp.fastsimon.com/mcp --scope user

--scope user makes it available in every project on your machine; drop it to add it for just the current directory.

Cursor / Windsurf / Cline / other MCP clients:

These use a JSON config instead of a command. Add the server with the same URL:

{
  "mcpServers": {
    "fastsimon": {
      "type": "http",
      "url": "https://mcp.fastsimon.com/mcp"
    }
  }
}

Where to put it:

  • Cursor — Settings → MCP → Add new MCP server (or edit ~/.cursor/mcp.json).
  • Windsurf — Settings → Cascade → MCP servers → Add (or edit ~/.codeium/windsurf/mcp_config.json).
  • Cline (VS Code) — the Cline panel → MCP Servers → Configure → paste the block.
  • Any other MCP client — add a streamable-HTTP (type http) server pointing at https://mcp.fastsimon.com/mcp.

That's it — no account, no keys. The MCP is read-only: it never sees or stores your project files (your assistant reads those locally), and it only serves content that is already public — this documentation and the published fast-simon-sdk type declarations.

Make your assistant use it every time​

Having the MCP installed is not the same as your assistant reaching for it. Assistants read their project instruction file on every session, so one short note there makes sure Fast Simon work goes through the MCP instead of the assistant's memory.

Add this to your project's CLAUDE.md (Claude Code) or AGENTS.md (Cursor, Windsurf, Codex and other assistants that read it):

## Fast Simon

- Before writing or changing any Fast Simon code, use the `fastsimon` MCP server:
  confirm every SDK method and field with `fastsimon_sdk_get_reference` instead of guessing.
- Only use documented APIs. If the MCP flags something as not part of the documented
  public API, do not build on it.
- A Fast Simon surface is not finished until its reporting events are wired up.
  Get them from `fastsimon_sdk_get_reference` and `fastsimon_sdk_get_event`.

Implement​

Ask in plain language, inside your storefront project:

Add a Fast Simon search results page to this theme.

Your assistant inspects your project, asks the MCP for the exact API shapes and the known-good recipe, and writes the code — adapted to your stack (Liquid theme, Hydrogen, headless, …). It loads the SDK from the CDN so your storefront always runs the latest published version.

Or start from a ready recipe — type / in Claude Code and pick one, e.g. search-result-page, collection-page, upsell-and-cross-sell-recommendations, visual-discovery-sample.

Review an existing integration​

Run the review prompt (type / and choose it, or just ask):

/review-integration

Your assistant reads your existing Fast Simon code and audits it against the MCP — confirming method signatures, flagging any use of undocumented/internal APIs, and checking that every surface fires its required reporting events (a page that renders but reports nothing is an incomplete integration). Every finding is grounded in a tool result, not a guess.

What it knows​

Ask your assistant about…The MCP provides
How to build a search / collection / recommendations / visual-discovery pageSearchable docs + ready recipes
The exact arguments a method takesThe authoritative signature from the SDK type declarations
Which events a surface must reportThe event constants and payloads for that surface
Whether something is safe to useA clear flag when a method/field is not part of the documented public API

Because that knowledge is derived from the SDK and docs themselves, it stays correct as the SDK evolves — your assistant won't invent a method or a parameter that doesn't exist.