How to Check an Invention for Novelty with AI
Before you file, find out what’s already out there. This tutorial checks an invention for novelty with AI with Patsnap Patent Research, from a technical disclosure to five compared prior-art references.
This workflow works in Claude, Claude Code, ChatGPT, or Codex when the selected client supports the required MCP connection. After connecting, the research steps and evidence boundaries are the same.
You’ve developed a new technical solution and need to know whether it looks novel before committing resources to a patent application. That means comparing your invention against millions of published patents and applications — something An AI assistant or coding agent can help you frame, but can’t retrieve or search on its own. Patsnap Patent Research supplies that live data through a unified MCP connection, running novelty searches directly from Claude and returning ranked candidate patents with feature-level comparisons — research leads for qualified IP review, not a final patentability opinion.
Prepare the invention disclosure
Novelty screening starts with a clear, complete technical description: the problem your invention solves, the key structural or functional elements, and how they work together — including materials, dimensions, and method steps where they distinguish your solution. A fuller description improves accuracy (the tool recommends a passage longer than 200 words), though shorter queries still run. Avoid marketing language and focus on the technical implementation; if your invention combines known components in a new way, describe that combination explicitly. Since the description becomes the search query, technical clarity matters more than prose style. Sketches, diagrams, or photos are optional and can improve retrieval, though text alone is usually enough.
Distinguish novelty screening from a patentability conclusion
A novelty search retrieves and compares prior art, helping you spot features that may distinguish your invention from what’s already published. It doesn’t render a legal opinion on patentability, validity, or freedom to operate — examination weighs novelty, inventive step, industrial applicability, and formal requirements, and this search addresses only the first at a preliminary level. Results depend on the description you supply, the database scope, and the ranking algorithm: a highly relevant reference can be missed if it uses different terminology or sits outside the search scope, and a reference flagged as similar may turn out to differ in ways that preserve novelty once examined closely. Treat results as research leads for a patent attorney or agent to interpret against claim-drafting strategy and examination practice in your target jurisdictions.
Connect Patsnap Patent Research
https://connect.patsnap.com/f176d7/mcp?apikey=yourapikey Replace yourapikey with your own Patsnap Open Platform API key before connecting. Keep the completed URL private. - Claude: add the remote MCP connection from Claude’s connector or integration settings.
- Claude Code: add the server as a remote HTTP MCP connection in your user or project configuration.
- ChatGPT: add the server through the supported connector or developer-mode MCP setup available to your workspace.
- Codex: add the server in Codex’s MCP configuration for the relevant local workspace.
The interface differs by client, but you use the same Patsnap connection URL and the same natural-language research prompt.
- Get a Patsnap Open Platform API key.
- Copy the connection URL or JSON config from the Connect panel on the Patsnap Patent Research page.
- Add the server to your AI assistant or coding agent or another MCP-compatible client, then restart it.
- Confirm novelty_search and get_task are available before you ask — the same server also exposes fto_review and design_fto for related workflows.
Start the novelty-search task
Check this invention for novelty: Thin aluminum inter-cell cooling plate with serpentine coolant channels, sealed phase-change-material pockets, inlet/outlet temperature sensing and delta-T flow control.Claude calls novelty_search with that description, lite mode for faster execution, and top_k set to 5 for five representative comparisons. The tool returns a task ID and runs asynchronously through seven steps — generating search queries, retrieving candidates, screening for relevance, comparing features, assessing novelty, and preparing a summary report. Claude polls get_task automatically and notifies you when it’s ready — typically a few minutes for a lite search. In the recorded run, task novelty-6ad9b23bec2e completed all seven steps.
What the search result gives you
50 candidates were retained after screening, with five core patents selected for detailed comparison:
| Publication | Covers |
|---|---|
| US20140272496A1 | Battery thermal-management system with integrated PCM and liquid cooling |
| US11223081B2 | Cooling plate with internal phase-change material chambers |
| US8039139B2 | Serpentine-channel liquid-cooling plate for battery modules |
| US10749225B2 | Thermal-management assembly with temperature-differential flow control |
| US10033072B2 | Hybrid thermal buffer combining active and passive cooling layers |
The report compared these against the disclosure and concluded the sealed PCM pockets integrated directly into the liquid-cooling plate appeared novel. Each reference showed partial overlap — serpentine channels in one, discrete PCM chambers in another, differential sensing in a third — but none combined all four features in the described configuration.
Refine around uncovered features
Use the returned patents to explore adjacent design space — Ask your AI assistant or coding agent to search for cooling plates combining metallic construction with embedded phase-change materials, or thermal-management systems that modulate flow rate on real-time temperature differentials. Each refinement reuses novelty_search with adjusted terminology. You can also supply a publication number from the initial result and Ask your AI assistant or coding agent to retrieve the full abstract, claims, or bibliographic details for deeper comparison. If the search surfaced close prior art, consider whether a design change, a narrower claim, or a different inventive-step argument might preserve patentability; if it found nothing blocking, expand the query to related materials, alternative implementations, or earlier filing dates to confirm the field is clear.
Prepare findings for qualified review
Compile the novelty report, compared patents, and any follow-up results into a summary for your patent attorney or IP advisor — highlight which features appeared in prior art and which combinations stayed uncovered, and note any terminology mismatches or jurisdictional gaps that might warrant a supplemental search. Provide the full technical description you used so the attorney can judge whether the search captured your invention’s complete scope, and flag any design changes made since so they can advise on whether a new search is needed before filing.
Check novelty before you file
Connect Patsnap Patent Research and compare your invention against published prior art.
Frequently asked questions
Can I just Ask your AI assistant or coding agent to check novelty without connecting Patsnap Patent Research?
No. An AI assistant or coding agent cannot retrieve or search current patent records on its own. It can help structure a disclosure and interpret patent language, but needs an MCP connection like Patsnap Patent Research to find and compare prior art.
What do I need to prepare before running a novelty search?
A clear technical description: the problem it solves, key structural or functional elements, how they work together, and any distinguishing materials, dimensions, or method steps. A passage over 200 words improves accuracy.
Does a novelty search result mean my invention is patentable?
No. It retrieves and compares prior art to help identify distinguishing features — it doesn’t render a legal opinion on patentability, which also weighs inventive step, industrial applicability, and formal requirements.
Can I refine my search after reviewing the initial results?
Yes. Adjust terminology or emphasis in a follow-up novelty search, or investigate a specific patent from the results by retrieving its full abstract, claims, or bibliographic details.
Can I use this workflow in Claude, Claude Code, ChatGPT, or Codex?
Yes, provided the selected client and workspace support the required MCP connection. Connect the same Patsnap server, then use the natural-language steps shown in this guide; interface details may differ by client.
Disclosure & disclaimer
Who published this
Published by Patsnap, which develops and sells the Patsnap Patent Research MCP server described here. This is an editorial tutorial, not an independent review.
How the information was gathered
Server capabilities and connection details are taken from the Patsnap Patent Research product page and Patsnap Open Platform documentation, as accessed August 2026. They are not the result of independent testing beyond the recorded novelty-search run described in this tutorial. Product capabilities change frequently and may have changed since publication.
Not legal advice
The novelty comparison in this tutorial is research support, not a patentability opinion. Consult qualified patent counsel before making a filing or claim-strategy decision.
Trademarks
All trademarks, service marks, product names, and company names mentioned — including Claude and Cursor — are the property of their respective owners and are used here solely for identification purposes. Their use does not imply any affiliation with, sponsorship by, or endorsement from their respective owners.