Card Grader MCP

Vision-based trading card identification, centering analysis, and grading assessment.

Supported card types: Pokémon TCG, Magic: The Gathering, Yu-Gi-Oh!, sports cards (Topps, Panini, Upper Deck, Bowman, etc.)

Grading services: CGC, PSA, BGS, SGC, TAG

---

Tools

Tool	Description	PPE
identify_card	Single front photo → name, set, rarity, centering, value	$0.03
identify_card_both_sides	Front + back → full centering both sides + damage	$0.05
analyze_centering	Centering-only analysis → L/R, T/B ratios, CGC/PSA/BGS sub-grades	$0.03
estimate_card_value	Text-only value lookup (no image needed)	$0.02
grade_assessment	Full submission assessment → submit/borderline/skip + reasoning	$0.04

---

Setup

Environment Variables

$export ANTHROPIC_API_KEY="sk-ant-..."

Set in Apify Console → Source → Environment Variables (mark as secret).

Local Development

cd card-grader
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
export ANTHROPIC_API_KEY="sk-ant-..."
python run_server.py

Server starts on http://localhost:4321/mcp

---

Testing

Encode a test image

$BASE64=$(base64 -w 0 /path/to/card.jpg)

Test identify_card

from server import identify_card
import base64

with open("card.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

result = identify_card(image_data=b64, mime_type="image/jpeg")
print(result)

Test analyze_centering

from server import analyze_centering
import base64

with open("card.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

result = analyze_centering(image_data=b64)
print(result["centering"])

Test estimate_card_value (no image)

from server import estimate_card_value

result = estimate_card_value(
    card_name="Charizard",
    card_set="Base Set",
    card_number="4/102",
    year=1999,
    rarity="holo rare",
    card_variant="shadowless",
    condition="near mint raw",
)
print(result)

Test grade_assessment

from server import grade_assessment
import base64

with open("card.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

result = grade_assessment(
    image_data=b64,
    target_grade="10",
    grading_service="CGC",
)
print(result["assessment"]["submission_recommendation"])
print(result["assessment"]["cgc_summary"])

Test identify_card_both_sides

from server import identify_card_both_sides
import base64

with open("front.jpg", "rb") as f:
    front_b64 = base64.b64encode(f.read()).decode()

with open("back.jpg", "rb") as f:
    back_b64 = base64.b64encode(f.read()).decode()

result = identify_card_both_sides(
    front_image_data=front_b64,
    back_image_data=back_b64,
)
print(result["card"]["cgc_summary"])

---

Deploy to Apify

rm -rf venv
apify push

After push: re-enable Actor Standby in Apify Console → Settings.

Claude Desktop config

{
  "mcpServers": {
    "card-grader": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://YOUR_USERNAME--card-grader.apify.actor/mcp?token=apify_api_XXXXX"]
    }
  }
}

---

CGC Centering Scale Reference

Sub-Grade	L/R & T/B
10 (Pristine)	55/45 or better
9.5 (Gem Mint)	60/40 or better
9 (Mint)	65/35 or better
8.5 (NM/Mint+)	65/35–70/30
8 (NM/Mint)	70/30–75/25
7.5 and below	Worse than 75/25

Note: Centering is measured from photos. Corners, edges, and surface require hands-on inspection for accurate sub-grades.

---

License

MIT — Mowell Labs 2025