TikTok for developers: search data without a scraper

TikTok for developers usually comes down to two bad options: an official API that only reads your own account, or a scraper you have to keep alive. You want the data for something else you are building, an agent that answers research questions, an internal dashboard, a pipeline that feeds a model, and you do not want a headless browser and a proxy pool in the middle of it. TokConnect is one HTTP endpoint with 26 tools behind a bearer key, and it returns something no scraper can reach: TikTok's own search-volume numbers.

What you'd ask on day one

Three things people do in the first hour. Every number below came out of these exact calls on 19 September 2026.

import asyncio
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

URL = "https://mcp.tokconnect.com/mcp"
HEADERS = {"Authorization": "Bearer ttk_YOUR_KEY"}

async def main():
    async with streamablehttp_client(URL, headers=HEADERS) as (read, write, _):
        async with ClientSession(read, write) as session:
            await session.initialize()
            tools = await session.list_tools()
            print(len(tools.tools), "tools")
            r = await session.call_tool(
                "search_topics",
                {"keyword": "raspberry pi projects", "limit": 10},
            )
            print(r.content[0].text)

asyncio.run(main())

{"queryId": "7668816613126193173",
 "queryText": "command line and Raspberry Pi projects",
 "popularity": 87,
 "searchVolume": 4657670,
 "videoCount": 0,
 "trend7d": [9495, 105547, 288955, 439580,
             1084257, 1601498, 4657670],
 "categoryL1": "Science and Technology",
 "categoryL2": "Digital"}

import json

async def walk(session, keyword, max_pages=2, count=10):
    offset, search_id = 0, None
    for _ in range(max_pages):
        args = {"keyword": keyword, "count": count,
                "sort": "relevance", "date": "any", "offset": offset}
        if search_id:
            args["search_id"] = search_id
        page = json.loads(
            (await session.call_tool("search_videos", args)).content[0].text
        )
        yield page["videos"]
        if not page["hasMore"]:
            return
        offset, search_id = page["nextOffset"], page["searchId"]

What it replaces

The pattern across that table: every one of them returns content. None returns demand. Creator Search Insights lives in the TikTok app and at tiktok.com/csi, and neither surface has an export, a history view or an API. That is the gap.

How it fits your week

1. Connect once. In Claude Code it is claude mcp add --transport http tokconnect https://mcp.tokconnect.com/mcp --header "Authorization: Bearer ttk_…". In Cursor, Codex or Gemini CLI it is a four-line block in the config file. Every client is on the connect page, and the Claude Code guide has the full command.
2. Ask in chat while you design. This is where the exploration happens: which keywords have volume, what the field names are, what an empty result looks like.
3. Move the calls you keep into code. Same endpoint, same tools, same key. The mcp Python package speaks Streamable HTTP, so a cron job is fifteen lines. There is no separate REST surface to learn and no second set of credentials.

Where the numbers come from: topic tools read TikTok's Creator Search Insights, and the video, creator, hashtag, sound and comment tools read ordinary TikTok search. Both are pulled live when you call them, not from a cache we refreshed last week. The 26 tools and what each returns are listed in the TikTok MCP guide.

What it does not do: no posting, no scheduling, no ads, no transcripts, no private or logged-in data. TokConnect is an independent product and is not affiliated with or endorsed by TikTok. If you need to publish to TikTok, that is the Content Posting API and a different problem, covered in the TikTok API map.

Straight answers

The plan that fits

Starter, $49 a month for 1,000 credits. Count your week in calls, because one call is one credit. A nightly job that runs search_topics across a dozen keywords is about 360 calls a month. Paging five pages of search_videos for four keywords every week adds 80. Comment pulls at 50 per page are two credits per hundred. That leaves most of the allowance for the questions you ask in chat while you are still deciding what to build. Start on Free, which is 100 credits and every tool unlocked, and move up only when the scheduled job outgrows it. Growth at $99 for 3,000 is the next step; the maths for both is on the pricing page.

Guides for developers and agent builders

• TikTok MCP: all 26 tools with their arguments and what each returns, plus how the community servers differ.
• TikTok API: the full map of official and unofficial access paths, so you can rule the wrong ones out fast.
• TikTok search API: search endpoints compared, with the quotas and a runnable example.
• TikTok scraper: what actually breaks in a self-hosted scraper, and what it costs to keep one running.
• Connect Claude Code: the one-line claude mcp add command and how to check the server is healthy.
• TikTok automation: which automated actions are allowed and which get an account banned.