ELDRICK Club Fitter on Grok

A starter for building a golf club fitter on Grok with the xAI API. Grok interviews the golfer, runs an ELDRICK fitting, shows the specs and top picks with store links, and finds a club fitter nearby.

Built on 100,000+ fittings.

ELDRICK is a remote MCP server at https://api.eldrick.com/mcp. xAI's API attaches remote MCP servers as a server-side tool: you name the server in the request, and xAI connects to it and calls its tools for you. Your code never calls ELDRICK directly. See xAI's Remote MCP Tools guide.

File	What it is
eldrick-mcp.json	The ELDRICK mcp tool (URL, label, allowed tools, attribution headers). Shared by both starters. No key in it.
system-prompt.md	The interview: club, handedness, speed, miss, then optional extras. Then fit, show, offer next steps.
python/grok_fitter.py	CLI chat on the xAI Python SDK (xai-sdk).
typescript/grok-fitter.ts	The same chat on xAI's OpenAI-compatible Responses API, with plain fetch.
.env.example	Every setting. Copy to .env.

Setup

1. Get an xAI API key at console.x.ai.
2. Get an ELDRICK developer key (eld_...) at api.eldrick.com. Every account includes 10 fittings a month; searching, scoring, comparing, grip sizing, bag gapping and finding fitters don't use one.
3. Put both in .env:

cp .env.example .env
# edit .env: XAI_API_KEY=...  ELDRICK_API_KEY=eld_...

Keys stay in the environment. They never go in source, in the system prompt or in a tool argument.

Run it

Python (3.10+):

cd python
python -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt
python grok_fitter.py

TypeScript (Node 20):

export XAI_API_KEY=... ELDRICK_API_KEY=eld_...
npx tsx typescript/grok-fitter.ts

Both take --print-config to show the exact tool they send (key redacted), with no network calls.

You: driver please
ELDRICK: Right- or left-handed?
You: right, about 98 mph, I slice it
  [calling analyze_fitting]
ELDRICK: (loft, shaft, length, lie and swing weight, then the top picks with store links) Fit by ELDRICK

How ELDRICK is attached

Python, xAI SDK (xai_sdk.tools.mcp; the SDK spells two fields allowed_tool_names and extra_headers):

import os
from xai_sdk import Client
from xai_sdk.chat import system, user
from xai_sdk.tools import mcp

client = Client(api_key=os.environ["XAI_API_KEY"])
chat = client.chat.create(
    model="grok-4.7",
    messages=[system(open("system-prompt.md").read())],
    tools=[mcp(
        server_url="https://api.eldrick.com/mcp",
        server_label="eldrick",
        authorization=os.environ["ELDRICK_API_KEY"],
        allowed_tool_names=["analyze_fitting", "get_fitting", "score_clubs", "build_store_link", "find_fitter"],
        extra_headers={"X-ELDRICK-Platform": "grok"},
    )],
)
chat.append(user("Fit me for a driver. Right-handed, 98 mph, I slice it."))
print(chat.sample().content)

Responses API (OpenAI-compatible: base_url="https://api.x.ai/v1" with the OpenAI SDK, or raw HTTP):

curl https://api.x.ai/v1/responses \
  -H "Authorization: Bearer $XAI_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4.7",
    "input": [{"role": "user", "content": "Fit me for a driver. Right-handed, 98 mph, I slice it."}],
    "tools": [{
      "type": "mcp",
      "server_url": "https://api.eldrick.com/mcp",
      "server_label": "eldrick",
      "authorization": "'"$ELDRICK_API_KEY"'",
      "headers": {"X-ELDRICK-Platform": "grok"}
    }]
  }'

Notes:

• authorization is the ELDRICK key. xAI puts it in the Authorization header of its calls to ELDRICK; ELDRICK accepts it with or without Bearer .
• allowed_tools keeps Grok to the tools you list. Leave it out and Grok sees every ELDRICK tool.
• X-ELDRICK-Platform: grok puts your traffic under Grok in ELDRICK's usage reports. Add X-ELDRICK-Traffic: test while you build, so test runs stay out of them. Neither header changes billing or limits.
• xAI documents remote MCP for the Responses API and the native SDK, not for the legacy Chat Completions endpoint.
• The model is grok-4.7 (xAI's current flagship as of 2026-10-09); set GROK_MODEL to change it.

Fittings and cost

analyze_fitting uses one of the account's fittings; every other tool in the starter is free. The system prompt tells Grok to send an idempotencyKey, wait on get_fitting when a fitting is still running, and retry a failed one with the returned key, so a retry is never charged twice. A fitting takes about 20 to 50 seconds.

Testing

npm test runs server/lib/grok-bot-template.test.ts, which checks this template offline: the tool matches xAI's documented mcp fields, names only tools ELDRICK serves, carries no key, and the Python starter prints the same config. It never calls xAI or ELDRICK. To also run one live call (a free search_clubs only, never a fitting):

XAI_API_KEY=... ELDRICK_API_KEY=eld_... ELDRICK_GROK_LIVE_TEST=1 \
  npx tsx --test server/lib/grok-bot-template.test.ts

Show the attribution

When you show a fitting, show "Fit by ELDRICK". Setup for every other assistant is at api.eldrick.com/connect.