tiktok-studio-mcp
MCP server for TikTok that publishes videos to your own TikTok account and retrieves video performance metrics through TikTok's official Content Posting and Display APIs.
README
tiktok-studio-mcp
MCP server for TikTok. Publishes videos to your own TikTok account and reads back how they performed, through TikTok's official Content Posting and Display APIs.
Built as the TikTok counterpart to
yt-studio-mcp.
What it can and cannot do
TikTok's creator APIs are narrower than YouTube's. Rather than advertise tools that cannot work, this server exposes only what the platform actually supports:
| capability | supported |
|---|---|
| Upload video (drafts or direct post) | yes |
| View / like / comment / share counts | yes |
| List your videos | yes |
| List or moderate comments | no — TikTok offers this only through the Research API, which is gated to academic institutions |
| Playlists, captions, thumbnails | no — no API surface |
Posting modes
TikTok gates direct publishing behind an app audit, so there are two modes,
selected with TIKTOK_MCP_MODE:
upload(default) — the video lands in your TikTok inbox/drafts and you finish posting it in the app. Works without an audit.publish— posts directly. Requires thevideo.publishscope, which needs TikTok's full app audit (2–4 weeks, demo video, privacy policy, domain verification).
Until the app is audited, TikTok forces everything an unaudited client posts
to private, whatever privacy_level you ask for. post_video says so in its
result rather than letting you assume something published.
Install
claude mcp add tiktok -s user -- uvx tiktok-studio-mcp
Authenticate
Register an app at developers.tiktok.com with
Login Kit, Content Posting API and Display API, requesting the scopes
user.info.basic, video.upload, video.list, video.publish. Set the
Desktop redirect URI to http://localhost:8902/.
export TIKTOK_MCP_CLIENT_KEY=...
export TIKTOK_MCP_CLIENT_SECRET=...
tiktok-studio-mcp auth
Prefer the environment over --client-key / --client-secret: command-line
arguments are visible to any user on the box via ps and land in shell history.
The flags still work as a fallback.
The flow is headless-friendly: it binds a fixed port and prints the consent URL rather than launching a browser. On a machine with a browser:
ssh -N -L 8902:localhost:8902 user@your-host
then open the printed URL.
Tools
| tool | purpose |
|---|---|
health_check |
credentials refresh, account reachable, granted scopes, active mode |
creator_info |
nickname, allowed privacy levels, duration cap, interaction toggles |
post_video |
upload a file; returns publish_id. Supports dry_run |
post_status |
poll a publish_id |
list_videos |
your videos, with statistics |
video_stats |
counts for specific video ids |
post_video calls creator_info first and validates privacy_level against
what the account actually allows, because TikTok rejects mismatches with an
unhelpful error.
Secrets
TIKTOK_MCP_SECRETS selects the backend: file (default,
~/.config/tiktok-studio-mcp/credentials.json, mode 600), env, or vaultproxy.
TikTok rotates refresh tokens — every refresh returns a new one and invalidates the old. This server writes the new token back on every refresh; if it did not, authentication would work for 24 hours and then fail with no obvious cause.
Limits
Enforced by TikTok, surfaced by this server:
- 6 requests per minute per access token on the init endpoints
- 5 pending shares per 24 hours
- chunked upload: files under 5 MB go whole; otherwise chunks are 5–64 MB, the final chunk may reach 128 MB, 1–1000 chunks, 4 GB maximum
Development
pip install -e '.[dev]'
ruff check src tests
pytest
Tests use an injected transport and need no TikTok credentials.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。