Claude.ai · Skills & Connectors · Tutorial
Build Your Own Clinical Research Skill for Claude
A step-by-step guide to creating a personal AI skill that taps into Clinical Trials and PubMed — simultaneously — from a single prompt.
Claude’s Skills system lets you teach Claude a repeatable, expert workflow — once — and invoke it on demand in any future conversation. Pair that with the Clinical Trials and PubMed connectors already built into Claude’s Connectors library, and you get a clinical research intelligence engine that can query both databases, synthesize findings, and produce a publication-grade report, all in one go.
This guide walks you through every step: connecting the data sources, prompting Claude to build and package the skill, saving it to your personal library, and then firing it up with both connectors live.
“Skills turn a brilliant one-off conversation into a repeatable, always-available workflow — no copy-pasting prompts, no re-explaining context.”
Connect Clinical Trials & PubMed
Before building the skill, make sure the two data connectors are enabled. These are Claude’s direct pipelines to ClinicalTrials.gov and the PubMed/NCBI literature database.
Open the Customize panel
Click the sidebar toggle or navigate to Customize → Connectors from the left rail.
↳ See Image 1Locate the Web connectors
Under the Web section you will find Clinical Trials (ClinicalTrials.gov) and PubMed listed as ready-to-connect sources.
Enable both
Click each connector to connect it. A blue toggle confirms the connection is active. These connectors give Claude live search access to both databases.
Ask Claude to Build the Skill
With the connectors live, start a conversation and describe the workflow you want Claude to codify. Be specific: name the databases, the output format, and the depth of analysis you expect.
Example prompt
“You have access to PubMed and ClinicalTrials.gov. I want you to build a skill that: searches both databases for AI/ML research in a given therapeutic area, summarises active trials, surfacing peer-reviewed publications, analyzes domain distribution and practitioner impact, and outputs a comprehensive publication-grade intelligence report with all source links. Capture this as a reusable skill.”
Claude will work through the logic, create the supporting files (SKILL.md and a references/report-template.md), and then package everything into a .skill file — as shown in Image 2. The skill description is automatically condensed for accurate triggering in future sessions.
Save the Skill to Your Personal Library
Once Claude finishes building the skill file, a save prompt appears at the top of the response. This is the single click that persists the skill across every future conversation.
Review the skill card
Claude surfaces a card showing the skill name (ai-clinical-research-intelligence), its type, and a download option.
Click “Save skill”
Hit the highlighted Save skill button in the top-right of the card. Claude saves the skill to your personal library immediately.
Confirm in Skills panel
Navigate to Customize → Skills → Personal skills. You’ll see ai-clinical-research-intelligence listed with its component files (SKILL.md and the references folder).
What’s saved
SKILL.md— the full workflow instructions Claude reads each time it’s triggeredreferences/report-template.md— the output structure for the intelligence report- Skill description — a condensed trigger phrase Claude uses to recognize when to apply the skill
Run the Skill with Both Connectors Active
Now comes the payoff. In any new conversation, activate the skill and both connectors together to get simultaneous queries across ClinicalTrials.gov and PubMed in a single prompt.
Open the attachment menu
Click the + button in the message bar to open the tools menu.
Enable the skill
Select Skills and toggle on /ai-clinical-research-intelligence. It appears as a blue tag in the prompt bar.
Enable both connectors
Select Connectors and toggle on Clinical Trials and PubMed. Both turn blue. Web search can remain active alongside these.
Send your research query
Type your topic — e.g. “Generate a clinical research intelligence report on AI in cardiology for 2025” — and send. Claude will follow the skill’s workflow, query both databases in parallel, and return a structured report.
What the Skill Produces
When triggered with both connectors live, the ai-clinical-research-intelligence skill returns a structured intelligence report covering:
Report sections
- Active and recently completed trials from ClinicalTrials.gov, with NCT IDs and sponsor info
- Peer-reviewed publications from PubMed, including DOIs and citation data
- Domain distribution breakdown — which specialties are publishing or trialing AI
- Identified benefits, risks, and limitations from the literature
- Practitioner impact summary — what clinicians and researchers should know
- All source links to official ClinicalTrials.gov and PubMed records
The same skill works for any therapeutic area or research question. Change the topic in your prompt and the workflow adapts — the skill handles the methodology so you can focus on the research question.
Tips for Getting the Most from Your Skill
Best practices
- Be specific in your research query: include the therapeutic area, year range, and desired output depth.
- If you want the skill to evolve, start a new conversation, request changes, and re-save — Claude will update the
SKILL.mdand overwrite the saved version. - You can have multiple personal skills active at once — pair this with a writing style skill for formatted outputs.
- The skill is private to your account; it won’t appear for other users.
- Skills work best when connectors are toggled on before sending the prompt — set them in the attachment menu before typing.
The ability to create personal skills that speak directly to specialized data sources marks a meaningful shift in how researchers can work with AI. Rather than re-prompting Claude from scratch for every literature review, you build once and deploy indefinitely — with full access to the live clinical evidence base baked in.
Try it with your own area of clinical interest. The methodology is the same; only the question changes.

