X posts and their replies. As JSON.
Posts from a list of AI researchers, labs and builders on X, each with the replies under it, collected a few times a day. Read-only, no key.
https://x.served.fyi/api/posts
30 requests a minute per IP · newest first · replies nested
Quick start
Ask for the latest posts. Every post comes back with its replies already attached.
curl "https://x.served.fyi/api/posts?limit=20"
# Just one account's own posts, without the replies
curl "https://x.served.fyi/api/posts?handle=simonw&kind=post&replies=0"import requests
page = requests.get("https://x.served.fyi/api/posts",
params={"limit": 20}, timeout=30).json()
for post in page["posts"]:
print(post["handle"], post["reply_count"], post["text"][:80])const res = await fetch("https://x.served.fyi/api/posts?limit=20");
const page = await res.json();
for (const post of page.posts) {
console.log(post.handle, post.reply_count, post.text.slice(0, 80));
}A post
One object per post from a followed account. Reposts and quotes point at the original.
{
"count": 1,
"next_cursor": "MjAyNi0xMS0xMFQwOToxMjowMCswMDowMHwxOTAwMDAwMDAwMDAwMDAwMDAx",
"posts": [{
"id": "1900000000000000001",
"kind": "quote",
"handle": "ada_builds",
"author_name": "Ada",
"account": "ada_builds",
"url": "https://x.com/ada_builds/status/1900000000000000001",
"text": "Ran the new model against our eval set. 3 regressions, all in tool calls.",
"published_at": "2026-11-10T09:12:00+00:00",
"first_seen_at": "2026-11-10T10:40:11+00:00",
"in_reply_to": null,
"repost_of": null,
"quote_of": {
"id": "1899999999999999990", "handle": "modelco",
"url": "https://x.com/modelco/status/1899999999999999990",
"text": "Model Co: Our new model is out today."
},
"media": [{ "type": "photo", "url": "https://pbs.twimg.com/media/example.jpg" }],
"links": ["https://example.com/evals/2026-11"],
"reply_count": 2,
"thread": { "fetches": 1, "last_fetched_at": "2026-11-10T11:30:02+00:00",
"next_fetch_at": "2026-11-11T09:12:00+00:00" },
"replies": [{
"id": "1900000000000000007", "kind": "reply", "handle": "lin_dev",
"text": "@ada_builds Which tool calls?",
"in_reply_to": { "id": "1900000000000000001", "handle": "ada_builds", ... },
"replies": [{
"id": "1900000000000000012", "kind": "reply", "handle": "ada_builds",
"text": "@lin_dev Mostly parallel ones. Writing it up.",
"replies": []
}]
}]
}]
}
idkindpost, quote, repost or reply. Top-level posts are never reply.handle, author_nameaccounthandle for top-level posts.textquote_of.text, not here.published_at, first_seen_atin_reply_to, repost_of, quote_of{id, handle, url} of the other post, or null. quote_of also has text.media, links{type, url, poster}, and outside links in the text. Empty lists when there are none.reply_countthreadnull for reposts, whose replies belong to the original.repliesreplies=0.html, rawfields=full: the rendered post and the item exactly as fetched, for anything the fields above miss.Replies
Each reply has the same fields as a post, plus its own replies, so a conversation nests as deep as it goes.
- A reply sits under the post or reply it answers, oldest first.
- If X didn't return the reply it answers, it sits at the top of the tree with
"parent_missing": true. - Replies are fetched twice: when a post is about 2 hours old and again at about 24 hours. After that the tree doesn't change.
- Each fetch gets the replies X shows first, up to about 40. Very busy posts will have more on X than here.
- A new post has
"replies": []until its first fetch.thread.next_fetch_atsays when that is.
Paging
Every response has a next_cursor. Send it back as cursor with the same filters to get the next, older page. When it is null, you have everything.
The cursor points at the last post you received, so new posts arriving while you page don't cause repeats or gaps.
import time, requests
url = "https://x.served.fyi/api/posts"
params = {"limit": 100, "since": "2026-11-01"}
posts = []
while True:
r = requests.get(url, params=params, timeout=30)
if r.status_code == 429: # over the rate limit
time.sleep(int(r.headers["Retry-After"]))
continue
page = r.json()
posts += page["posts"]
if not page["next_cursor"]:
break
params["cursor"] = page["next_cursor"]
print(len(posts), "posts")const url = new URL("https://x.served.fyi/api/posts");
url.searchParams.set("limit", "100");
const posts = [];
while (true) {
const res = await fetch(url);
if (res.status === 429) { // over the rate limit
await new Promise(r => setTimeout(r, res.headers.get("Retry-After") * 1000));
continue;
}
const page = await res.json();
posts.push(...page.posts);
if (!page.next_cursor) break;
url.searchParams.set("cursor", page.next_cursor);
}# Page 1
curl "https://x.served.fyi/api/posts?limit=20"
# Page 2: same request, plus the next_cursor from page 1
curl "https://x.served.fyi/api/posts?limit=20&cursor=MjAyNi0xMS0xMFQwOToxMjow..."Reference
Everything is GET and returns JSON.
/api/posts /api/posts/{id} /api/accounts /api/status /healthz{"ok": true}. Not rate limited.Parameters for /api/posts
limit1–200cursornext_cursor from the previous page.handlehandle=simonw,swyx. Not case-sensitive.kindpost, quote or repost, comma-separated. kind=post,quote leaves out reposts.since, untilsince and before until. A date (2026-11-10) or a date and time; UTC unless you give an offset.qrepliesreplies=0 leaves the reply trees out, for smaller, faster pages.fieldsfields=full adds html, raw, last_seen_at, source and conversation_id.Errors
400{"error": "since must be an ISO date or datetime"}.404429Retry-After header, also in the body as retry_after_seconds.503