scraper.py

# pylint: skip-file

import time
import re
import md5
import requests
import json

INSTAGRAM_URL = "https://www.instagram.com"
HASHTAG_ENDPOINT = "/graphql/query/?query_hash={}&variables={}"
USER_AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/65.0.3325.181 Safari/537.36"

def get_first_page(hashtag):
    return requests.get(INSTAGRAM_URL + "/explore/tags/{}/".format(hashtag), headers={"user-agent": USER_AGENT})

def get_csrf_token(cookies):
    return cookies.get("csrftoken")

def get_query_id(html):
    script_path = re.search(r'/static(.*)TagPageContainer\.js/(.*).js', html).group(0)
    script_req = requests.get(INSTAGRAM_URL + script_path)
    return re.findall('return e.tagMedia.byTagName.get\\(t\\).pagination},queryId:"([^"]*)"', script_req.text)[0]

def get_rhx_gis(html):
    return re.search(r'rhx_gis":"([^"]*)"', html).group(1)

def get_end_cursor_from_html(html):
    return re.search(r'end_cursor":"([^"]*)"', html).group(1)

def get_end_cursor_from_json(json_obj):
    return json_obj['data']['hashtag']['edge_hashtag_to_media']['page_info']['end_cursor']

def get_params(hashtag, end_cursor):
    return '{{"tag_name":"{}","first":50,"after":"{}"}}'.format(hashtag, end_cursor)

def get_ig_gis(rhx_gis, params):
    return md5.new(rhx_gis + ":" + params).hexdigest()

def get_posts_from_json(json_obj):
    edges = json_obj['hashtag']['edge_hashtag_to_media']['edges']
    return [o['node'] for o in edges]

def get_posts_from_html(html):
    json_str = re.search(r'window._sharedData = (.*);</script>', html).group(1)
    json_obj = json.loads(json_str)
    graphql = json_obj["entry_data"]["TagPage"][0]["graphql"]
    return get_posts_from_json(graphql)

def make_cookies(csrf_token):
    return {
        "ig_pr": "2",
        "csrftoken": csrf_token,
    }

def make_headers(ig_gis):
    return {
        "x-instagram-gis": ig_gis,
        "x-requested-with": "XMLHttpRequest",
        "user-agent": USER_AGENT
    }

def get_next_page(csrf_token, ig_gis, query_id, params):
    cookies = make_cookies(csrf_token)
    headers = make_headers(ig_gis)
    url = INSTAGRAM_URL + HASHTAG_ENDPOINT.format(query_id, params)
    req = requests.get(url, headers=headers, cookies=cookies)
    req.raise_for_status()
    json_obj = req.json()
    end_cursor = get_end_cursor_from_json(json_obj)
    posts = get_posts_from_json(json_obj['data'])
    return posts, end_cursor

def scrape_hashtag(hashtag, sleep=3):
    """
    Yields scraped posts, one by one
    """
    first_page = get_first_page(hashtag)
    csrf_token = get_csrf_token(first_page.cookies)
    query_id = get_query_id(first_page.text)
    rhx_gis = get_rhx_gis(first_page.text)
    end_cursor = get_end_cursor_from_html(first_page.text)
    home_posts = get_posts_from_html(first_page.text)

    for post in home_posts:
        yield post

    while end_cursor is not None:
        params = get_params(hashtag, end_cursor)
        ig_gis = get_ig_gis(rhx_gis, params)
        posts, end_cursor = get_next_page(csrf_token, ig_gis, query_id, params)
        for post in posts:
            yield post
        time.sleep(sleep)


# main
for post in scrape_hashtag("summer"):
    print post['id']
    # do stuff

@Kingson that would be only for the first page, right?

Thanks a ton, saved a lot of time.
I modified it slightly to scrape instagram user's public timeline photos link.
Could you please explain "ig_pr":2 in cookies, I don't find it necessary, thanks.
Also you could change while True: to while end_cursor != None: in case the scraper reaches the very last page(where end_cursor is None).

@ketankr9 thanks for the suggestions. ig_pr was needed at some point, so I'm going to leave it there just in case..
I also integrated the loading of posts from the home, thanks about that.

Is anybody else getting a 500 status after roughly 70 posts? The same script worked fine 2 weeks ago ... Maybe they tightened their temporary IP blocking?!?

Has somebody had problems with the scraper?

Basically the cookies are not having anymore the csrf token, and so, it is breaking