Define what needs screening before launch

A cross-border launch exposes your product to local design-patent rights that a text or keyword search won’t surface — design protection is visual, so the relevant prior art is often only discoverable through image comparison. This MCP server runs a multi-stage workflow — image feature extraction, similarity ranking, feature comparison, and report generation — and surfaces tiered risk alerts for each comparison. The output is a reviewable list of potentially relevant design rights, not a legal clearance opinion; it fits between initial product photography and formal freedom-to-operate counsel.

Prepare the image and search scope

The workflow needs one public HTTPS product image URL showing the design you plan to launch — a photograph or rendering that clearly presents the shape, ornamentation, and proportions a design examiner or competitor would notice. Cropped catalog shots, exploded assembly views, or glamour angles with heavy shadows reduce similarity accuracy.

You’ll also set one or more country codes (US, CN, EP, JP, KR) to scope the search, a legal-status filter (active, pending, or both — passed as numeric codes, comma-separated), and an optional filing-date range. An optional free-text field describing the product can improve retrieval, and a separate structured field accepts Locarno classification entries if you have them.

For this tutorial, the input is a public smartwatch product image, filtered to US and EP, restricted to active and pending design patents, in English.

1Connect the MCP server to Claude

1. Get a Patsnap Open Platform API key.
2. Copy the connection URL or JSON config from the Connect panel on the Design Patent Infringement Search page.
3. Add the server to Claude Desktop or another MCP-compatible client, following that client’s setup steps.
4. Confirm the three tools — submit_workflow, get_task_status, get_stage_result — are available before you ask.

2Submit an image-led similarity search

Ask Claude to screen your product image for similar design patents in your chosen markets — for example: “Screen this smartwatch image for design-patent risk in the US and EP, active and pending patents only: [HTTPS image URL].” Claude calls submit_workflow with your image and filters, and the server queues the multi-stage comparison and returns a task ID. Claude then polls get_task_status until the workflow finishes — 142.68 seconds in the recorded run — and calls get_stage_result to retrieve the comparisons. You don’t need to type any parameters or JSON yourself; describing the image and your filters in plain language is enough.

3Review the similarity comparisons

The recorded run returned 16 comparisons with zero processing failures:

The similarity score reflects visual overlap identified by the image-comparison algorithm — it doesn’t measure claim scope, ornamental-versus-functional boundaries, or the legal test for design-patent infringement. A feature-comparison view highlights which portions of the design drove the ranking.

4Refine without a new image

Once the workflow completes, ask Claude to retrieve detailed feature-comparison data for specific results, filter the table by jurisdiction or a similarity threshold, or generate a summary report — each follow-up reuses the task ID and pulls further detail from the completed workflow rather than issuing a fresh image query.

What this looks like in code

Everything above happens through plain conversation — you never need to write this yourself. This section is for readers who want to see the mechanics Claude is calling behind the scenes.

{
  "url": "https://example.com/images/smartwatch-product.jpg",
  "country": ["US", "EP"],
  "simple_legal_status": "1,2",
  "main_field": "smartwatch with circular face and rotating bezel"
}

country takes the jurisdiction codes to screen; simple_legal_status uses numeric codes — 0 expired, 1 active, 2 pending — comma-separated; main_field is an optional free-text description, not a classification code, that helps the retrieval step.

{
  "task_status": "completed",
  "processing_time_seconds": 142.68,
  "comparisons_returned": 16,
  "results": [
    {
      "publication_number": "USD933065S",
      "similarity_score": 0.96,
      "risk_signal": "High"
    }
  ]
}

This output shape reflects the confirmed test-run data described above, not a full published API schema — field names beyond what’s documented are illustrative.

Escalate relevant results for specialist review

The workflow delivers ranked, jurisdiction-specific design-patent leads in minutes, but it doesn’t replace verification by IP counsel. High-similarity scores and feature callouts guide internal triage; a patent attorney reviews the actual drawings, applies the correct infringement test per jurisdiction, confirms current legal status in the official register, and assesses whether design-around options exist. Use the structured comparison data to brief your legal team efficiently.