Why label questions need exact wording, not a summary

A question like “what does the label say” isn’t asking for background knowledge — it’s asking for the precise text filed with the regulator. Generic language models synthesize answers from training data and can’t distinguish an approved label section from a literature summary someone else wrote about it. When the wording itself is the answer — for a safety claim, a dosing instruction, a pregnancy warning — a paraphrase isn’t good enough, and the source has to be the live label, not a memory of one.

Why this needs Regulatory & Guidelines

Regulatory & Guidelines runs semantic search across the full FDA label collection and returns matched text chunks with homology scores, so you can describe a concept — “mechanism of action,” “pharmacokinetics” — and retrieve the relevant section even when your phrasing differs from the label’s own heading. The same server searches clinical guidelines with the same approach, so one connection covers both “what’s approved” and “what’s recommended” questions.

Prepare your query

State the drug name, the therapeutic concept, or both — “semaglutide” retrieves label sections for that active ingredient across approved products, while “semaglutide mechanism of action” focuses on mechanism sections specifically. One constraint worth knowing: the search rejects meta-words like “literature,” “paper,” “review,” or “patent” — stick to biology, chemistry, and pharmaceutical terms, since the tool is built for content search, not document-type filtering. For this tutorial, the input is query “semaglutide” with language “EN”.

1Connect Regulatory & Guidelines

1. Get a Patsnap Open Platform API key.
2. Copy the connection URL or JSON config from the Connect panel on the Regulatory & Guidelines page.
3. Add the server to your AI assistant or coding agent or another MCP-compatible client, then restart it.
4. Confirm the FDA label and clinical guideline search tools are available before you ask.

2Run the example query

What does the FDA label for semaglutide say about its mechanism and pharmacokinetics?

Claude recognizes the request and calls fda_label_search(query="semaglutide", lang="EN"), which runs a vector similarity search against the label database and returns the top-matching text chunks ranked by semantic similarity.

3Read the returned evidence

Each chunk is tagged with the product name — Wegovy, in this case — and a homology score, with the highest-scoring matches appearing first. Because these are excerpts from the official label, the wording is the approved regulatory text, not a paraphrase — the semantic search finds it even when your query’s phrasing differs from the label’s own section heading.

4Go further with follow-up questions

From the same connection, Ask your AI assistant or coding agent to:

• “Show me only the dosing and administration section for semaglutide.”
• “What does the label say about semaglutide drug interactions?”
• “Compare the mechanism language in the semaglutide label to the liraglutide label.”
• “Search clinical guidelines for semaglutide use in type 2 diabetes management.” — pivots from approved label text to consensus treatment recommendations using clinical_guideline_search, without changing tools.