I
IvanBBaev/tiktok-mcp
Unofficial TikTok MCP server — policy-governed tools over the official Display and Content Posting APIs, with…
在 GitHub 查看 ↗链接已于 2026-11-12 核查 · HTTP 200 · 打开的是厂商自己的官网
安装与配置
客户端配置（取自该仓库的安装命令）
{
"mcpServers": {
"ivanbbaev-tiktok-mcp": {
"command": "npx",
"args": ["tiktok-mcp-ai"]
}
}
}解析自我们于 2026-11-07 抓取的该仓库 README；我们未执行过。
command/args 来自仓库写明的安装命令；不确定时请以该项目 README 为准。
VS Code（.vscode/mcp.json —— 键名是 servers）
{
"servers": {
"ivanbbaev-tiktok-mcp": {
"command": "npx",
"args": ["tiktok-mcp-ai"]
}
}
}VS Code 顶层键是 <code>servers</code>，其余与上面那份相同。
Claude Code（命令行）
claude mcp add --transport stdio ivanbbaev-tiktok-mcp -- npx tiktok-mcp-ai
Claude Code 不收 JSON 文件，用这条命令直接加。
配置项的键名以你的客户端为准：客户端与 CLI
- ★ 数取值
- 1 ★ · IvanBBaev/tiktok-mcp · 取自 GitHub 搜索 API · （2026-11-13 20:00:50 核对）
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型：开源项目
- 收录与活跃
- 收录于 2026-11-08 · 最近见到 2026-11-14 · 共见到 7 次 · 被 1 个来源列出
- 评分构成
- 综合分 103 · 1 ★ · 1 个来源 · repo
- 安装
npx tiktok-mcp-ai
常见问题（答案都来自我们的记录）
怎么安装？
该仓库给出的安装命令是
npx tiktok-mcp-ai。我们的记录里它是仓库自己提供的；不保证在你的客户端上直接可用。它适合谁 / 用来做什么？
我们把它归在这些标签下: 开源项目。标签来自条目自身，不是我们强加的分类。
谁维护它、还在更新吗？
仓库
IvanBBaev/tiktok-mcp；我们收录于 2026-11-08；最近见到 2026-11-14。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。# tiktok-mcp-ai — Unofficial TikTok MCP Server
以下摘录自该仓库的 README（片段） · 在 GitHub 读完整 README →
📖 Documentation site →
An unofficial MCP (Model Context Protocol)
server for TikTok — a safe, policy-governed tool surface over the official
TikTok for Developers APIs: the Display API
(read a creator's profile and videos) and the Content Posting API (publish
videos and photo carousels). It adds OAuth 2.0 (PKCE) user authorization,
plan-then-execute write safety, an append-only publish journal, multi-account
profiles, and token-efficient results — so a model can work with TikTok without
knowing any of the platform's sharp edges (24-hour tokens, cursor pagination,
chunked uploads, per-token rate limits, the app-audit gate, and an error
envelope that hides failures inside HTTP 200 responses).
> Unofficial project. This is an independent, community-built MCP server. It
> is not affiliated with, endorsed by, or sponsored by TikTok or ByteDance
> Ltd. "TikTok" is a trademark of its owner, used here only nominatively to
> identify the service this server connects to. See Trademark.
Contents: Quick demo · Features ·
Requirements · Setup ·
Configure credentials · Run / debug ·
Develop · Tools · Security notes ·
Uninstall · Project documentation ·
Support · Trademark
_Built and maintained in my own time — if it helps, a
GitHub Sponsors tip keeps it going.
Full Support options are near the end._
## Quick demo
The signature move is safe publishing: a write tool never mutates on the
first call. Point your MCP client at a TikTok app (Setup), authorize
once (npx tiktok-mcp-ai login), and ask:
1. "Post this clip — but show me exactly what will happen first." The model
calls tiktok_post_video without a plan_id. Nothing is posted: the result
is a full, human-readable preview — the resolved creator identity, the
privacy level, every flag, the chunk plan for the upload, the exact upstream
request that would be sent — plus a single-use plan_id valid for 10 minutes:
jsonc
// tiktok_post_video — preview (no plan_id ⇒ nothing is posted)
"source": "file",
"file_path": "clips/launch-teaser.mp4",
"post_info": {
"title": "Launch day! #buildinpublic",
"privacy_level": "PUBLIC_TO_EVERYONE",
},
2. "Looks right — do it." The model calls the same tool again with the
plan_id from the preview. The payload is re-resolved through the identical
code path, its SHA-256 digest is compared against the plan, the plan is consumed
atomically, and only then is the upload dispatched — so a retry can never
double-post, and nothing but the previewed payload can ever run:
jsonc
// tiktok_post_video — execute (the plan_id selects execution)
"source": "file",
"file_path": "clips/launch-teaser.mp4",
"post_info": {
"title": "Launch day! #buildinpublic",
"privacy_level": "PUBLIC_TO_EVERYONE",
},
"plan_id": "plan_9f2c7b1e4a6d0c83f5b2e19a7d4c6f80",
3. "Did anything actually post?" Every attempt — intent before the request,
outcome after — is written to a local append-only journal. Read it back with a
tool, no matter which client made the post:
bash
npx tiktok-mcp-ai doctor # among other checks: reconciles the publish journal
All read tools are safe to call anytime; the four write tools are the only ones
gated by plan-then-execute, and they return the publish_id immediately so the
An unofficial MCP (Model Context Protocol)
server for TikTok — a safe, policy-governed tool surface over the official
TikTok for Developers APIs: the Display API
(read a creator's profile and videos) and the Content Posting API (publish
videos and photo carousels). It adds OAuth 2.0 (PKCE) user authorization,
plan-then-execute write safety, an append-only publish journal, multi-account
profiles, and token-efficient results — so a model can work with TikTok without
knowing any of the platform's sharp edges (24-hour tokens, cursor pagination,
chunked uploads, per-token rate limits, the app-audit gate, and an error
envelope that hides failures inside HTTP 200 responses).
> Unofficial project. This is an independent, community-built MCP server. It
> is not affiliated with, endorsed by, or sponsored by TikTok or ByteDance
> Ltd. "TikTok" is a trademark of its owner, used here only nominatively to
> identify the service this server connects to. See Trademark.
Contents: Quick demo · Features ·
Requirements · Setup ·
Configure credentials · Run / debug ·
Develop · Tools · Security notes ·
Uninstall · Project documentation ·
Support · Trademark
_Built and maintained in my own time — if it helps, a
GitHub Sponsors tip keeps it going.
Full Support options are near the end._
## Quick demo
The signature move is safe publishing: a write tool never mutates on the
first call. Point your MCP client at a TikTok app (Setup), authorize
once (npx tiktok-mcp-ai login), and ask:
1. "Post this clip — but show me exactly what will happen first." The model
calls tiktok_post_video without a plan_id. Nothing is posted: the result
is a full, human-readable preview — the resolved creator identity, the
privacy level, every flag, the chunk plan for the upload, the exact upstream
request that would be sent — plus a single-use plan_id valid for 10 minutes:
jsonc
// tiktok_post_video — preview (no plan_id ⇒ nothing is posted)
"source": "file",
"file_path": "clips/launch-teaser.mp4",
"post_info": {
"title": "Launch day! #buildinpublic",
"privacy_level": "PUBLIC_TO_EVERYONE",
},
2. "Looks right — do it." The model calls the same tool again with the
plan_id from the preview. The payload is re-resolved through the identical
code path, its SHA-256 digest is compared against the plan, the plan is consumed
atomically, and only then is the upload dispatched — so a retry can never
double-post, and nothing but the previewed payload can ever run:
jsonc
// tiktok_post_video — execute (the plan_id selects execution)
"source": "file",
"file_path": "clips/launch-teaser.mp4",
"post_info": {
"title": "Launch day! #buildinpublic",
"privacy_level": "PUBLIC_TO_EVERYONE",
},
"plan_id": "plan_9f2c7b1e4a6d0c83f5b2e19a7d4c6f80",
3. "Did anything actually post?" Every attempt — intent before the request,
outcome after — is written to a local append-only journal. Read it back with a
tool, no matter which client made the post:
bash
npx tiktok-mcp-ai doctor # among other checks: reconciles the publish journal
All read tools are safe to call anytime; the four write tools are the only ones
gated by plan-then-execute, and they return the publish_id immediately so the
收录徽章（可选）
想把收录标记放进 README 就复制这行，可选，不影响收录。
[](https://anclum.com/zh/s/ivanbbaev-tiktok-mcp/)
- 来源
- github · https://github.com/IvanBBaev/tiktok-mcp
- 按领域浏览
- 开源项目
- Traction
- 1 ★
- 收录于
- 今天
同类推荐
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。