Drugs.com Drug-Interaction Check

Purpose

Given a list of two or more drugs (generic, brand, or autocomplete-derived ddc_ids), return every flagged drug-drug, drug-food/lifestyle, and drug-condition (disease) interaction surfaced by the Drugs.com Interaction Checker, with severity (major | moderate | minor | unknown), the two parties of the interaction, a patient-facing summary, the professional/clinical rewrite (mechanism + management), and the canonical Drugs.com URL. Honors an optional condition filter (e.g. pregnancy, breastfeeding, kidney disease, liver disease). Read-only — never click Save List, Print, Email, Sign Up, or submit any personal-health form. Output is informational only and must include a disclaimer that it is not medical advice.

When to Use

A patient or clinician-facing agent that needs to check a multi-drug regimen before answering "is it safe to take X with Y?"
Pre-fill screening for a triage flow: surface every major drug-drug interaction and any condition-specific warnings for a known patient condition.
Bulk-checking a medication list against a single new prescription.
Anywhere you would otherwise scrape the rendered Interaction Checker page — the URL-param + JSON-search path is ~50× cheaper and avoids the full JS-rendered UI.

Workflow

Drugs.com's official API is licensee-only, but the public Interaction Checker is backed by two public, unauthenticated JSON/HTML endpoints that together cover the entire task. Use those directly via bb fetch --proxies and skip browser driving entirely. The page-driven autocomplete + click-Add flow is documented as a fallback only.

Network constraint observed in this sandbox: CDP WebSocket connections to connect.*.browserbase.com are DNS-filtered, so the entire skill was implemented via Browserbase Fetch (bb fetch --proxies). The Fetch API rides through residential proxies and Akamai-passes without any session cookies. If you are running in an environment with full Browserbase access, you can substitute live browse open calls — the same URLs work — but you almost never need to.

1. Resolve every input drug to `(ddc_id, brand_name_id)`

Drugs.com identifies a drug in the Interaction Checker by a pair of integers: ddc_id (the ingredient/concept ID) and brand_name_id (0 if you want the generic; non-zero for a specific brand SKU). Resolve each input string via the public interaction-search endpoint:

bb fetch --proxies \
  "https://www.drugs.com/api/interaction/search/?search=$(printf '%s' "$NAME" | jq -sRr @uri)"

Response shapes (application/json):

// Exact match — usually one drug
{ "type": "FOUND",   "search": "ibuprofen",
  "drugs": [{ "type": "GX", "ddc_id": 1310, "brand_name_id": 0, "name": "ibuprofen" }] }

// Brand → generic mapping
{ "type": "FOUND",   "search": "lipitor",
  "drugs": [{ "type": "BN", "ddc_id": 276, "brand_name_id": 128,
              "name": "Lipitor", "generic_name": "atorvastatin" }] }

// Ambiguous partial match — multiple candidates returned (typeahead state)
{ "type": "PARTIAL", "search": "tylen",
  "drugs": [ { "type": "BN", "ddc_id": 11, "brand_name_id": 12, "name": "Tylenol", ... },
             { "type": "BN", "ddc_id": 11, "brand_name_id": 2701, "name": "Tylenol Extra Strength", ... },
             ... ] }

// Not found
{ "type": "UNKNOWN", "search": "xyznotreal", "drugs": [] }

Selection rules:

On FOUND with one entry → use that pair.
On FOUND/PARTIAL with multiple entries → prefer the exact case-insensitive match of name against the input. If none, prefer type: "GX" (generic) over type: "BN" (brand). If still tied, ask the caller to disambiguate; do not silently pick the first result.
On UNKNOWN → surface a resolution_failed entry in the output for that input; do not include it in drug_list.

The pair is encoded as "{ddc_id}-{brand_name_id}" in the URL — e.g. 1310-0 (generic ibuprofen), 276-128 (brand Lipitor).

2. Fetch the interaction report

The Interaction Checker accepts a comma-separated list of pairs as the drug_list query parameter. Fetch twice in parallel — once for the consumer narrative, once for the professional/clinical rewrite — and merge per block:

LIST="1310-0,2311-0"   # ibuprofen, warfarin
CONSUMER_URL="https://www.drugs.com/interactions-check.php?drug_list=$LIST"
PRO_URL="https://www.drugs.com/interactions-check.php?drug_list=$LIST&professional=1"

bb fetch --proxies "$CONSUMER_URL" > consumer.json
bb fetch --proxies "$PRO_URL"     > professional.json

Both URLs return a single ~170 KB HTML page. The ?professional=1 toggle rewrites every interaction-reference block's body from the patient-facing summary to the clinical GENERALLY AVOID: / MANAGEMENT: / <details class="ddc-reference-list"> form. The professional view does not contain the consumer text — that is why both are needed for the task's full output schema.

A bb fetch --proxies round-trip is ~1–2 s on these pages; both can be fired in parallel.

3. Parse the report HTML

The report is structured around four <h2> section headers that appear in this exact order (any of them may be absent if the section is empty):

`<h2>` text	Block type
`Interactions between your drugs`	drug-drug pairs
`Therapeutic duplication warnings`	same-class duplicates (no severity badge — text-only)
`Drug and food/lifestyle interactions`	drug × food, alcohol, tobacco, vitamin K, etc.
`Drug and disease interactions`	drug × medical condition

Within each section, every interaction is a <div class="interactions-reference"> whose first child is:

<div class="interactions-reference-header">
  <span class="ddc-status-label status-category-{major|moderate|minor|unknown}">Major</span>
  <h3>
    {drug1} <svg class="ddc-icon-drugvsdrug">…</svg> {drug2-or-condition-or-food}
  </h3>
  <p>Applies to: {comma-joined drug names}</p>
</div>

…followed by one or more <p> paragraphs of body text (consumer narrative in the default view; GENERALLY AVOID: / MANAGEMENT: clinical prose in the ?professional=1 view) and, on the professional view, a <details class="ddc-reference-list"> block with literature citations.

Block delimitation: split each section's HTML on every occurrence of <div class="interactions-reference">. The block runs until the next such opening div, or until the next <h2> (whichever comes first). Do not rely on matching </div> — drugs.com nests the closing tags inside the references-list <details> so a naive close-on-</div> regex over-consumes.

Pull out:

severity: capture group of status-category-([a-z]+) (lowercase).
pair_label: text of <h3>, stripped of the SVG and whitespace.
applies_to: comma-list inside <p>Applies to: …</p>.
summary / professional_detail: HTML→text of the body paragraphs.
references: optional, only in the professional view, parsed from the <details> <li>s.

Merge consumer and professional blocks by their (section, pair_label, applies_to) key — the order of blocks within a section is identical across the two fetches.

4. Build the per-pair canonical URL

For drug-drug interactions, the canonical detail URL is:

https://www.drugs.com/drug-interactions/{a_name}-with-{b_name}-{a_ddc}-{a_bn}-{b_ddc}-{b_bn}.html

a_name and b_name are the lowercase canonical names from step 1 sorted alphabetically — and Drugs.com canonicalizes the URL by alphabetical name even when your input order is different. Example: input (simvastatin 2067-0, clarithromycin 685-0) produces clarithromycin-with-simvastatin-685-0-2067-0.html, not simvastatin-with-…. Sort the pair before building the URL so it matches the <link rel="canonical"> on the report page.

For food/lifestyle and condition blocks there is no per-block detail URL — return the report-page URL (/interactions-check.php?drug_list=…) plus the pair_label so the caller can navigate manually.

5. Optional `condition` filter

When a condition is supplied (pregnancy, breastfeeding, kidney disease, liver disease, etc.), the report already lists every drug × condition pairing under Drug and disease interactions. Filter the parsed drug-condition blocks by case-insensitive substring of the condition against the pair_label (e.g. condition: "liver disease" matches warfarin Liver Disease and ibuprofen Liver Disease). Do not add ?condition= or similar to the URL — there is no such parameter; the disease section is always returned in full.

Pregnancy/breastfeeding queries are special-cased: Drugs.com surfaces those primarily on the per-drug pregnancy page (/pregnancy/{drug}.html) rather than as a disease interaction. For pregnancy or breastfeeding, also fetch https://www.drugs.com/pregnancy/{ddc_name}.html for each resolved drug and extract the <dt>FDA Pregnancy Category</dt> and the lactation summary if needed. (Out of scope of the basic flow; document as a "if condition ∈ {pregnancy, breastfeeding}, also do X" branch in the caller.)

6. Aggregate the output

Build the summary block from per-severity counts across all three block types and set recommend_consult_prescriber: true whenever any major is present. Always include the disclaimer string:

"This output is informational only and is not medical advice. Consult a licensed
healthcare provider before changing any medication regimen."

Browser fallback

If bb fetch --proxies is blocked for any reason — typically a transient Akamai 403 — drive the page UI instead:

SID=$(bb sessions create --keep-alive --verified --proxies | jq -r '.id')
browse --connect "$SID" open "https://www.drugs.com/interaction/list/"
# For each drug:
browse --connect "$SID" click '#livesearch-interaction'
browse --connect "$SID" type "ibuprofen"
browse --connect "$SID" wait timeout 1500            # let autocomplete dropdown render
browse --connect "$SID" click 'role=option[name="ibuprofen"]'
browse --connect "$SID" click 'button:has-text("Add")'
# After adding ≥ 2 drugs, the page lands on the same report HTML as the URL-param path.

The form action is <form action='/interaction/list/' method='post'>. The browser path costs ~10× the network roundtrips of the URL-param path and produces the same HTML — only use it when the API path fails.

Site-Specific Gotchas

READ-ONLY. Never click Save list, Print, Email, Sign Up, or any control inside the <form id="drug-interactions-search"> block beyond what's needed to populate the list in the browser fallback. The /api/interaction/list-save/ endpoint is CSRF-protected and would mutate a logged-in user's saved lists if a session cookie were present.
Drugs.com canonicalizes drug order in the URL by alphabetical name. drug_list=2067-0,685-0 (simvastatin first) renders the same page as 685-0,2067-0 (clarithromycin first), and both <link rel="canonical"> to clarithromycin-with-simvastatin-685-0-2067-0.html. When building the per-pair detail URL in your output, sort the pair so it matches the canonical.
ddc_id is per-ingredient, brand_name_id is per-SKU. Use brand_name_id=0 for the generic. Multiple brand SKUs share a ddc_id: Tylenol, Tylenol Extra Strength, and Tylenol Arthritis Pain are all ddc_id=11 (acetaminophen) with different brand_name_ids (12, 2701, 2953). Combination products like Tylenol PM live under a different ddc_id entirely (64, acetaminophen/diphenhydramine).
/api/interaction/search/ returns the truth; /api/autocomplete/?type=interaction does not. The latter endpoint exists but always returns {"resultCount":0,"categories":[]} without an active page-context CSRF token. Always use /api/interaction/search/?search=… for resolution.
/api/interaction/list-drugs/ and /api/interaction/list-names/ require a CSRF token. They respond {"error":{"code":403,"message":"Forbidden","details":"Invalid CSRF token"}} for any out-of-band call. They are not needed — the report HTML at /interactions-check.php?drug_list=… contains everything those endpoints would have returned.
?professional=1 rewrites the body of every interaction block. The default view (consumer) and ?professional=1 view (clinical) have the same block count and order, but their <p> bodies are mutually exclusive. To produce both summary and professional_detail in one record, fetch both URLs and zip the blocks. Do not assume the professional view will contain consumer text — it does not.
Severity enum is major | moderate | minor | unknown. The first three were observed across our test pairs (ibuprofen+warfarin, simvastatin+clarithromycin, caffeine+tramadol, metformin+lipitor+ibuprofen, acetaminophen+vitamin C). unknown is documented by Drugs.com but did not surface in any tested combination; emit unknown when the status-category- class is present but the value is anything other than the three observed.
Single-drug drug_list returns 302 to /drug-interactions/{drug}.html (a single-drug summary page, not a list report). Always enforce ≥ 2 distinct ddc_ids before fetching. Two SKUs of the same generic (e.g. 11-12,11-2701 — two Tylenol forms) also resolve to a single-drug page; collapse duplicate ddc_ids before building the list.
"No drug ↔ drug interactions were found" is a valid empty result, not a parse failure. Detect it by looking for the literal string "No drug ⇄ drug interactions were found" inside the Interactions between your drugs H2 section, or — equivalently — by finding zero <div class="interactions-reference"> inside that section's wrapper. The food and disease sections are almost never empty for any real drug.
Disease section is large. A two-drug pair routinely produces 50–100 drug-condition blocks because every drug in the list expands against every known condition contraindication. The agent's output should chunk or summarize this; do not return the full disease list unfiltered to a low-context UI.
The naïve <div class="interactions-reference">…</div> regex over-consumes. Drugs.com nests the closing tags inside the references-list <details> block, so block N's closing </div> is several <details> deep. The robust delimiter is the next opening <div class="interactions-reference"> or the next <h2> — split on those, not on the closing tag.
/api/, /drug-interactions-all/, /interaction/list/, /interactions-list-drugs.php, /search.php, /search-wildcard-phonetic.php are all Disallow:'d in robots.txt. The fetches we make are still served (200 with full content) because Drugs.com gates by user-agent string, not by URL path matching at the edge — but a polite agent should keep traffic low. Stay ≤ 1 RPS sustained and use a Verified UA via bb fetch --proxies.
Akamai is present but lenient on the Interaction Checker pages. A residential proxy (--proxies) is enough; --verified was not needed for the Fetch API path. If you switch to live browser driving, add --verified because the login/save flows are more aggressively challenged.
Stop at the report page. Do not paginate, do not click into per-drug detail pages, do not follow See also: recommendations. The report contains every datum the schema requires.
pregnancy and breastfeeding are not "diseases" in Drugs.com's taxonomy. They surface on per-drug pages at /pregnancy/{drug}.html, not on the Interaction Report. When the caller passes condition: "pregnancy" or condition: "breastfeeding", fetch the per-drug pregnancy page in addition to the interaction report. Other conditions (kidney disease, liver disease, hypertension, …) are filterable from the disease section directly.
The disclaimer is non-optional. Drugs.com's own footer reads "This material is provided for educational purposes only and is not intended for medical advice, diagnosis or treatment." Pass this through to your output; do not paraphrase or omit.

Expected Output

{
  "disclaimer": "This output is informational only and is not medical advice. Consult a licensed healthcare provider before changing any medication regimen.",
  "input_drugs": [
    {
      "input": "Advil",
      "resolved": {
        "type": "brand",
        "ddc_id": 1310,
        "brand_name_id": 782,
        "name": "Advil",
        "generic_name": "ibuprofen",
        "active_ingredients": ["ibuprofen"],
        "drug_list_token": "1310-782"
      }
    },
    {
      "input": "warfarin",
      "resolved": {
        "type": "generic",
        "ddc_id": 2311,
        "brand_name_id": 0,
        "name": "warfarin",
        "generic_name": "warfarin",
        "active_ingredients": ["warfarin"],
        "drug_list_token": "2311-0"
      }
    }
  ],
  "report_url": "https://www.drugs.com/interactions-check.php?drug_list=1310-782,2311-0",
  "canonical_url": "https://www.drugs.com/drug-interactions/ibuprofen-with-warfarin-1310-0-2311-0.html",
  "condition_filter": null,
  "interactions": {
    "drug_drug": [
      {
        "severity": "major",
        "drugs": ["ibuprofen", "warfarin"],
        "applies_to": ["ibuprofen", "warfarin"],
        "summary": "Using warfarin together with ibuprofen may increase the risk of serious bleeding complications, especially in the gastrointestinal tract…",
        "professional_detail": "GENERALLY AVOID: Nonsteroidal anti-inflammatory drugs (NSAIDs) may potentiate the hypoprothrombinemic effect and bleeding risk associated with vitamin K antagonists… MANAGEMENT: NSAIDs should be administered with vitamin K antagonists only if benefits are expected to outweigh the increased risk…",
        "mechanism_class": ["pharmacodynamic", "additive"],
        "references_count": 34,
        "url": "https://www.drugs.com/drug-interactions/ibuprofen-with-warfarin-1310-0-2311-0.html"
      }
    ],
    "drug_food": [
      {
        "severity": "moderate",
        "drug": "warfarin",
        "counterpart": "food/lifestyle (vitamin K intake)",
        "summary": "Nutrition and diet can affect your treatment with warfarin. Keep your vitamin supplement and food intake steady…",
        "professional_detail": null,
        "mechanism_class": ["pharmacokinetic"],
        "url": "https://www.drugs.com/interactions-check.php?drug_list=1310-782,2311-0#warfarin-food-lifestyle"
      },
      {
        "severity": "moderate",
        "drug": "ibuprofen",
        "counterpart": "ethanol (alcohol)",
        "summary": "Ask your doctor before using ibuprofen together with ethanol. Do not drink alcohol while taking ibuprofen…",
        "professional_detail": null,
        "mechanism_class": ["additive"],
        "url": "https://www.drugs.com/interactions-check.php?drug_list=1310-782,2311-0#ibuprofen-food-lifestyle"
      }
    ],
    "drug_condition": [
      {
        "severity": "major",
        "drug": "ibuprofen",
        "condition": "Asthma",
        "summary": "NSAIDs are contraindicated in patients with history of asthma, urticaria, or other allergic-type reactions…",
        "professional_detail": "NSAIDs are contraindicated in patients with history of asthma, urticaria…",
        "mechanism_class": ["pharmacodynamic"],
        "url": "https://www.drugs.com/interactions-check.php?drug_list=1310-782,2311-0#ibuprofen-asthma"
      }
    ],
    "therapeutic_duplication": []
  },
  "summary": {
    "drug_drug_count": 1,
    "drug_food_count": 4,
    "drug_condition_count": 69,
    "by_severity": { "major": 49, "moderate": 25, "minor": 0, "unknown": 0 },
    "has_major_interactions": true,
    "recommend_consult_prescriber": true
  }
}

Edge-case shapes:

// Empty drug-drug section (e.g. acetaminophen + vitamin C)
{ "interactions": { "drug_drug": [], "drug_food": [...], "drug_condition": [...] },
  "summary": { "drug_drug_count": 0, "has_major_interactions": false, ... } }

// One or more inputs unresolved
{ "input_drugs": [
    { "input": "xyznotreal", "resolved": null, "resolution_error": "UNKNOWN" },
    { "input": "ibuprofen",  "resolved": { ... } } ],
  "interactions": null,
  "error": "fewer_than_two_resolvable_drugs" }

// Ambiguous input (PARTIAL with multiple distinct ddc_ids)
{ "input_drugs": [
    { "input": "tylen", "resolved": null, "resolution_error": "AMBIGUOUS",
      "candidates": [
        { "ddc_id": 11, "brand_name_id": 12,   "name": "Tylenol", "generic_name": "acetaminophen" },
        { "ddc_id": 11, "brand_name_id": 2701, "name": "Tylenol Extra Strength", "generic_name": "acetaminophen" },
        { "ddc_id": 64, "brand_name_id": 8542, "name": "Tylenol PM", "generic_name": "acetaminophen/diphenhydramine" }
      ] } ],
  "error": "ambiguous_drug_name" }

// Single-drug list (rejected before fetch)
{ "error": "fewer_than_two_distinct_ddc_ids" }