TikTok search API: every working option in 2026

There is no public TikTok search API. TokConnect is the practical route for developers and agents: one API key searches TikTok videos, accounts, hashtags and sounds, and also returns TikTok's own search counts and TikTok Shop sales estimates, inside ChatGPT, Claude and Cursor through its MCP server or from your own code. A free key works with the MCP server; every paid plan adds a REST API (API docs). The same data powers TokConnect's research agent.

TikTok's only official keyword search is the Research API's video query, open to approved academic and non-profit researchers at 1,000 requests a day. REST vendors such as HasData and SociaVault resell raw search results.

All three routes are below, with real responses from the third: a search for "mechanical keyboard" on 19 September 2026 returned a top video with 16,593,119 plays.

Share or embed this infographic

Free to use on your site, in a deck or a newsletter with a credit link to TokConnect. Copy the code below.

<a href="https://tokconnect.com/guides/tiktok-search-api/"><img src="https://tokconnect.com/graphics/tiktok-search-api/summary.png" alt="No public TikTok search API. Here is what works instead: 1 official keyword search, the Research API, academics only; 1,000 Research API requests a day; 16.6M plays on the top “mechanical keyboard” result; 7.0B lifetime views on #mechanicalkeyboard" width="1200" height="630" style="max-width:100%;height:auto"></a><br>Source: <a href="https://tokconnect.com/guides/tiktok-search-api/">TikTok Search API: Every Working Option in 2026</a> on TokConnect

Copy embed codeDownload PNG

What does TikTok officially offer?

Four official API families matter for search; see every official TikTok API. Only one of them searches.

The video query endpoint is a good API. You POST a query object with conditions on keyword, hashtag_name, username, region_code, create_date, music_id, effect_id, video_id and video_length, and you get back up to 100 records with view, like, comment and share counts, plus voice_to_text. Paging uses cursor plus the search_id from the first response, and has_more tells you when to stop. The quota is 1,000 requests a day, up to 100,000 records, reset at midnight UTC, which search engines showed from TikTok's FAQ on 5 October 2026.

The catch is the application. You must work at a qualifying academic institution or non-profit, have a defined research proposal, and commit to non-commercial use. If you are building a product, you will not be approved. The Research API guide covers the process.

The unofficial APIs

Four vendors show up for this query, plus one open-source library, and TokConnect, the MCP route in the next section, sits in the table with them. The four vendors and the library work the same way underneath: they call TikTok's own web and app endpoints from their infrastructure, sign the requests, rotate proxies, and sell you the result as REST.

Checked 5 October 2026 on each vendor's own page, except TikAPI. HasData's price was checked 8 October. HasData, Apify and TokConnect prices are compared per 1,000 results in HasData vs Apify vs TokConnect.

Cost is per call and adds up on discovery work, where you throw most results away. Stability matters more: these are private endpoints, so shapes change without a changelog and your parser breaks on a Tuesday. The open-source route removes the bill and doubles the maintenance.

Get a free API key

A TikTok search API you can call without approval

The third route is TokConnect's MCP server. Instead of REST endpoints you get tools, so an agent (ChatGPT, Claude, Codex, Cursor, Gemini CLI) can call them in a conversation, and a script can call the same tools with one key (see Calling it from Python below).

TokConnect's endpoint is https://mcp.tokconnect.com/mcp and the header is Authorization: Bearer YOUR_TOKCONNECT_KEY. Keys are free and come from the API key page with a Google sign-in. You start with 50 free credits, then paid plans run from $34.99 a month billed yearly (500 credits a month); one credit buys one lookup, and failed lookups are free. Four of its search tools are shown below, each with a real response.

Get a free API key

video_search (v2: search_videos)

video_search takes keyword, limit (1 to 20), sort (relevance, most_liked or newest) and date (any, day, week, month, half_year). Asking for five results on mechanical keyboard on 19 September 2026 returned this, trimmed to one video:

{
  "videos": [
    {
      "id": "7594781996459642126",
      "desc": "Gray-black gradient keyboard!☠️🎮 #keyboard #customkeyboard #mechanicalkeyboard ...",
      "authorUniqueId": "aa***",
      "playCount": 16593119,
      "diggCount": 1124785,
      "commentCount": 5507,
      "shareCount": 82214,
      "collectCount": 154763,
      "createTime": 1768298093,
      "durationMs": 14934
    }
  ],
  "count": 4,
  "hasMore": true,
  "nextOffset": 5,
  "searchId": "2026091903345638DB5976135F38977E2C",
  "keyword": "mechanical keyboard",
  "sort": "relevance",
  "date": "any"
}

TikTok video search via TokConnect. Open the snapshot, which has both pages with media URLs stripped.

Note count came back as 4 when 5 were asked for. Treat the requested size as a ceiling, not a promise, and read the returned count.

The pagination contract

This is the part people get wrong. Video search paging needs two values together, not one.

• nextOffset is the offset to send on the next call. Page one returned 5.
• searchId is the session token. Send it back as search_id on any call where offset is above 0.
• hasMore says the upstream response advertises another page.
• Each page returns its own searchId. Page two of our run came back with 2026091903345880A67D9C27A1599AE7B2, a different value.

With the v2 search_videos, send offset: 5 with search_id from page one and you get page two, three more videos, nextOffset: 10. Send the offset without the search ID and you get an empty session error, not results. Keep the keyword, sort and date identical across pages too.

creator_search, hashtag_search, demand_keywords

creator_search (v2: search_users) returns accounts with follower counts, video counts, total likes and the verification flag. The first hit for the same keyword:

{"uid": "6764046294925100037", "uniqueId": "mechanicalkeyboards",
 "nickname": "Mechanical Keyboards", "followerCount": 6505,
 "followingCount": 66, "awemeCount": 189, "totalFavorited": 24549,
 "verified": true}

hashtag_search (v2: search_hashtags) returns lifetime totals for the tag, which is the number TikTok shows on the hashtag page:

{"id": "48797542", "name": "mechanicalkeyboard",
 "viewCount": 7022731596, "videoCount": 374361, "useCount": 374361}

demand_keywords (v2: search_suggestions) is TikTok's own autocomplete, in TikTok's own order. It is quick keyword research, because these are phrases people actually type:

["mechanical keyboard", "mechanical keyboard sound",
 "mechanical keyboard for gaming", "mechanical keyboard 75%",
 "mechanical keyboard 60%", "mechanical keyboard and mouse",
 "Mechanical keyboard wireless", "mechanical keyboard asmr",
 "mechanical pencil", "mechanical keyboard amazon"]

Creators, hashtags and autocomplete for one keyword. Open the snapshot. More on mining autocomplete in the search suggestions guide.

product_top_selling: which keyboards sell

Search tells you which keyboard videos get played. product_top_selling tells you which keyboards sell. On 20 October 2026 it ranked the KiiBOOM Phantom98 Lite first for "mechanical keyboard" on TikTok Shop US: an estimated $86.32K from 804 units in the 30 days to 18 October, sold by 155 creators. The next two, the Solakaka A99 Pro and the AULA S75 Pro, made about $39K and $37K.

Data: TokConnect, TikTok Shop US, estimated revenue and units, 30 days to 18 October 2026. Open the snapshot.

Calling it from Python

Any MCP client works. With the official mcp package, a paged search is about twenty lines:

import asyncio, json, os
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

URL = "https://mcp.tokconnect.com/mcp"
AUTH = {"Authorization": f"Bearer {os.environ['TOKCONNECT_KEY']}"}

async def search(keyword, pages=2, limit=20):
    async with streamablehttp_client(URL, headers=AUTH) as (r, w, _):
        async with ClientSession(r, w) as s:
            await s.initialize()
            args = {"keyword": keyword, "limit": limit, "sort": "relevance"}
            videos = []
            for _ in range(pages):
                page = json.loads(
                    (await s.call_tool("video_search", args)).content[0].text)
                videos += page["videos"]
                if not page.get("next"):
                    break
                args = page["next"]  # the arguments for the following page
            return videos

for v in asyncio.run(search("mechanical keyboard")):
    print(f'{v["playCount"]:>10,}  @{v["authorUniqueId"]}  {v["desc"][:60]}')

import httpx2
from mcp.client.streamable_http import streamable_http_client

async with streamable_http_client(URL, http_client=httpx2.AsyncClient(headers={"Authorization": f"Bearer {KEY}"})) as (read, write):
    async with ClientSession(read, write) as session:
        ...

Install with pip install mcp. The same call in an agent needs no code at all, which is usually the point of this route. The TikTok MCP guide has the client configs.

Search TikTok videos for "mechanical keyboard", two pages sorted by relevance. Then pull TikTok's autocomplete for the same term. Give me a table of the ten videos with views and handle, and tell me which autocomplete phrases the top videos are not covering.

Which route should you pick?

Answer three questions and the choice makes itself.

• Do you want an agent or a script to answer questions about TikTok? Use TokConnect. Search, comments, hashtags, Creator Search Insights topics and TikTok Shop sales estimates come through one connection, with no parser to maintain.
• Do you need results in your own pipeline? Use TokConnect from code: the endpoint takes JSON-RPC over HTTP with a bearer key, as in the Python example above.
• Are you an academic with a proposal? TikTok's Research API is free and official, but closed to commercial use.

Comparing MCP servers rather than REST vendors? Nine TikTok MCP servers compared covers HasData's, Apify's, Bright Data's and TokConnect's side by side.

Pick TokConnect if you want TikTok search results, TikTok's own search counts and TikTok Shop sales estimates in an agent or a script today. Publishing needs TikTok's Content Posting API, and our TikTok scraper guide covers page scraping. Comments have their own cursor contract.

How we got these numbers

The TikTok responses on this page were pulled through TokConnect's MCP server: two pages of search_videos (5 per page, relevance order), plus search_users, search_hashtags and search_suggestions for the keyword "mechanical keyboard". Play, like, follower and hashtag counts are TikTok's public totals on that day, global; hashtag views and video counts are lifetime totals for the tag. Media URLs and secUids were stripped before saving, and video authors are masked.

Vendor facts come from each vendor's own documentation and pricing pages. TokConnect's tool count (34) is the server's current tool list, and the keyboard sales figures come from one product_top_selling call (keyword "mechanical keyboard", 30 days) on 21 October 2026. Research API limits come from TikTok's developer FAQ, which search engines showed with these figures on 5 October 2026 while the page itself returned an error to us. Data: video search snapshot, users, hashtags and autocomplete snapshot, HasData facts, tool catalog.