Study the pattern.
Make your own.
Install the Sandcastles connector, analyze videos, compare patterns and design a small content test.
Research Better Content with Claude and Sandcastles
The video promises the setup for turning saved social video links into research. Sandcastles documents a Claude desktop plugin with social-media analysis skills, including /analyze. Availability, supported sources and usage may change with the service. Use the tool to learn from examples, then create original work rather than reproducing someone else’s script or footage.
One useful step at a time.
- 01Connect the tool
- 02Collect examples
- 03Analyze patterns
- 04Run a test
Connect and test one video
Open the Sandcastles MCP setup page, confirm the current plan requirements and follow its desktop installation flow. Read the connector’s access and data terms before authorizing it. Start with one public video you are allowed to study.
Use the connected Sandcastles tool to analyze this video: [URL]. Report the topic, opening hook, format, structure, pacing and visible call to action. Separate observed elements from inferred reasons it may have performed well.
The service’s own page lists /analyze and related watchlist skills. Use the exact skill name shown in your installed version.
Compare a small, relevant set
Collect five to ten public examples from the same niche and audience. Record creator, date, platform, length, topic, hook and observable engagement. A viral example can be unusual or boosted by an existing audience, so do not infer causation from view count alone.
Compare these [number] analyzed videos. Make a table of recurring hook types, formats, topics and calls to action. Include counterexamples. Suggest three original content ideas for my audience [description] that use the underlying pattern without copying wording or visual assets.
Turn findings into a test
Choose one variable to test, such as hook structure or video length. Keep the audience and topic as similar as practical. Write two original scripts, publish them through your normal process, and compare a metric tied to your goal. Save what you learned, including ideas that did not work.
Build a two-week content experiment from these findings. Choose one variable, draft three original hooks, define one primary metric and a simple review date. Do not promise that any format will go viral.
Research worksheet
| Field | Record |
|---|---|
| Niche and intended audience | Who the video serves |
| Hook | The first sentence or visual question |
| Format | Talking head, screen demo, story, carousel, etc. |
| Proof | What evidence appears in the clip |
| Action | What the viewer is asked to do |
| Hypothesis | Why it might work, marked as a guess |
| Original test | Your own version and how you will measure it |
Sources & notes
Product details were checked on 5 October 2026. Menus, plans, and integrations can change. The video is the source of the lead-magnet promise; the links below support the practical instructions and corrections in this guide.
- Sandcastles MCP: Plugin setup and social research skills.
- Claude skills, connectors and plugins directory: How to browse extensions.
- Use skills in Claude: Skill availability and setup.
The clip states “14 skills”; the installed catalog may change. Check the vendor’s current page.