素材面板菜单新增「下载更多素材」直达入口;收口 unicut-mcp 服务等未提交改动

- 上传素材旁下拉菜单新增第三项「下载更多素材」,新窗口打开素材下载文档站 doc.88531.cn/docs/mrfx/mrfx-1g99qum57n5qk(zhCN/enUS 文案 my_download_more,Download 图标)
- sourceImport:showDirectoryPicker 不可用时降级,非 FS 浏览器可直接进编辑器
- 收口今日未提交改动:unicut-mcp 服务与 nginx/compose 反代、README/CHANGELOG 更新日志
This commit is contained in:
3970260761 committed 2026-11-11 19:45:54 +08:00
1 parent 5b8e53f9c8
commit f2ac228849
15 files changed
+1104 -5

No files matched your search

+3
View File
@@ -27,6 +27,9 @@ ssh
public/local-ai.json
.env
# MCP 服务生成产物(工程 ZIP / 配音,运行时数据,勿提交)
mcp-out/
# 数据质量检查的生成报告(每次运行都会变,不入库)
data-quality/reports/
+22
View File
@@ -6,6 +6,28 @@
## 2026-11-11
### ✨ 素材面板菜单
**素材面板新增「下载更多素材」直达入口**
- 上传素材右侧的下拉菜单(在线录音 / 文字转语音)新增第三项「下载更多素材」,点击在新窗口打开素材下载文档站 `https://doc.88531.cn/docs/mrfx/mrfx-1g99qum57n5qk`
- 文案入语言表:`zhCN.my_download_more` / `enUS.my_download_more`,图标用 `Download`(@icon-park/react),菜单项渲染与既有项同构
### 🤖 MCP 接入
**新增 `unicut-mcp` 服务:AI 客户端直接调用米剪能力(Model Context Protocol)**
- 新增 `mcp/` 服务(Python FastMCP,Streamable HTTP):抓取文章 `grab_url`(与前端 `extractFromHtml` 同规则)、口播文案 `generate_script`(OpenAI 兼容网关)、配音 `tts` / `list_voices`(unicut-tts)、工程生成 `create_video_project`、文件夹模式工程读写 `list/read/write_project`、自检 `service_status`
- `create_video_project` 产出与米剪导出同构的工程 ZIP(`project.json + materials/{type}/`,过 `checkNdata` 九键门禁):图片校验对齐前端(≥160×120 / ≤10MB)、有配音自动铺满时长、文案按字数加权切字幕,元素结构照抄官方 demo 工程真实字段
- nginx 同源反代 `https://<站点>/mcp`(SSE 关缓冲长连接)与 `https://<站点>/mcp-out/*`(工程 ZIP 下载);compose 新增 `unicut-mcp` 服务,密钥走项目 `.env`(`MCP_LLM_KEY` 等,不入库)
- 文档:`mcp/README.md`(工具明细 + Claude Desktop / Claude Code / Cursor / Hermes 接入配置 + docker run / compose 两种部署),主 README「AI」章新增 MCP 小节与更新表行
- 版本号 `1.0.0 → 1.0.1`(`package.json` / README 发布 tag 同步;授权中心基线仍为 1.0.0,客户端上报 1.0.1 判定「已是最新」)
### 🐛 修复(/grab/ 抓取)
- 公众号图片防盗链:`/grab/` 向上游清空 `Referer`(带异源 Referer 时 `mmbiz.qpic.cn` 只给 140×140 降级缩略图,无 Referer 才给原图,实测 2KB vs 47KB)
- 正文根节点优先 `#js_content`:微信文章页没有 `article`/`main`,旧选择器会退化到整页 `body` 混进账号栏/留言噪音
### ✨ 新功能
**AI 一键出片 · 素材三选一(图片 / Word·PDF 文档 / 网址)**
+36 -2
View File
@@ -53,9 +53,11 @@
| 模板导入修复 | 修复"导入第一个模板后,再导入会把浏览器拖崩"的问题,支持连续换模板 |
| 品牌与首页 | 产品名改为米剪,首页新增导航站 / 开源地址入口,移除收费栏目与账号类外链 |
| 接入授权中心 | 版本号、公告、检查更新(授权中心不可达时静默降级,绝不拦功能) |
| MCP 接入 | 新增 `unicut-mcp` 服务:Claude / Cursor / Hermes 等 AI 客户端经 `https://<站点>/mcp` 调用抓取文章、生成口播、edge-tts 配音、生成米剪工程 ZIP |
### 最近更新(2026-11-11)
- **素材面板「下载更多素材」入口**:上传素材旁的下拉菜单(在线录音 / 文字转语音)新增第三项,点击新窗口直达素材下载文档站 https://doc.88531.cn/docs/mrfx/mrfx-1g99qum57n5qk;
- **AI 一键出片素材三选一**:图片 / Word·PDF 文档(自动提取正文与配图,扫描件有提示)/ 网址自动抓取正文和配图(同源 `/grab/` 反代,拒绝内网回环目标);纯文字素材自动补标题封面;
- **纯文本模型也能出片**:出片卡片新增提示「只需纯文本模型即可使用;模型支持多模态(视觉)就能自动生成出片」;未配视觉模型时剥掉图片块、按图片编号循环分配画面,纯文本主模型不再被顶回错误;
- **断网 lint 门禁修复**:`external_refs` 20 → 0(PDF.js 命名空间入册、示例域名清除、官方自有域名注明理由入白名单),构建门禁恢复全绿;
@@ -188,6 +190,20 @@ UNICUT 是一款面向创作者、开发者与企业团队的在线视频编辑
> [!IMPORTANT]
> AI 对话、AI 一键出片、AI 生图 / 生视频默认接入官方 AI 网关(`ai.88531.cn`),在编辑器右上角「配置模型」粘贴一个 API Key 即可解锁,对话 `mimo-v2.6-flash`、视觉 `deepseek-v4-flash-vision-exp`、生图 `flux-schnell`、生视频 `inkling` 均已预填。配音(本机 edge-tts)、抠图、字幕、导出等能力不消耗任何外部额度。
### MCP:让 AI 客户端直接调用米剪
米剪内置 [MCP](https://modelcontextprotocol.io)(Model Context Protocol)服务 `unicut-mcp`,站点同源端点 `https://<站点>/mcp`。接入后,Claude Desktop、Claude Code、Cursor、Hermes 等 AI 客户端可以把米剪当工具用:
| 工具 | 能力 |
| --- | --- |
| `grab_url` | 抓网页正文与配图(公众号文章可用,规则与编辑器内抓取一致) |
| `generate_script` | 大模型生成短视频口播文案 |
| `tts` / `list_voices` | edge-tts 多音色配音(本机免费) |
| `create_video_project` | 图片 + 文案 + 配音 → 生成米剪工程 ZIP,米剪「工程文件」一键导入即可出片 |
| `list_projects` / `read_project` / `write_project` | 工作文件夹模式下的工程读写 |
典型工作流:*AI 抓文章 → 写口播 → 配音 → 生成工程 ZIP → 你在米剪导入*。部署、客户端接入配置与工具明细见 [mcp/README.md](./mcp/README.md)。
## 🧰 技术实现
米剪采用 React + TypeScript 构建编辑器界面与业务层,通过 MobX 管理复杂编辑状态;画布由 PixiJS / WebGL 实时渲染,音视频解码、处理和导出由 Web Worker、FFmpeg WebAssembly、MediaBunny 等能力协同完成。素材文件可写入用户授权的本地目录,工程元数据和文件句柄通过 Dexie / IndexedDB 管理。
@@ -214,7 +230,7 @@ docker run -d --name unicut -p 3006:80 -p 3005:443 \
gitea.88531.cn/jinghui/unicut:latest
```
打开 `http://服务器IP:3006` 即用。Registry 地址 `gitea.88531.cn` 免登录可拉;镜像与源码构建产物完全一致,每次发版同步更新 `latest` 与 `v1.0.0` 两个 tag。
打开 `http://服务器IP:3006` 即用。Registry 地址 `gitea.88531.cn` 免登录可拉;镜像与源码构建产物完全一致,每次发版同步更新 `latest` 与 `v1.0.1` 两个 tag。
### 方式二:Docker Compose部署(源码 compose 构建)
@@ -246,6 +262,23 @@ services:
volumes:
- ../unicut-tts/audio:/data/audio
# MCP 服务(Model Context Protocol):AI 客户端经 https://<站点>/mcp 调用米剪能力
# (抓取正文配图 / 大模型口播文案 / edge-tts 配音 / 生成米剪工程 ZIP)。
# 密钥与站点地址从项目 .env 注入(MCP_LLM_KEY 等,勿提交仓库)。
unicut-mcp:
build: ./mcp
container_name: unicut-mcp
restart: always
environment:
- MCP_LLM_BASE=${MCP_LLM_BASE:-http://192.168.68.19:3099/v1}
- MCP_LLM_KEY=${MCP_LLM_KEY:-}
- MCP_LLM_MODEL=${MCP_LLM_MODEL:-mimo-v2.6-flash}
- MCP_PUBLIC_URL=${MCP_PUBLIC_URL:-https://127.0.0.1:3005}
- MCP_PROJECTS_DIR=${MCP_PROJECTS_DIR:-}
volumes:
# 生成的工程 ZIP / 配音 mp3 落盘(nginx /mcp-out/ 提供下载)
- ./mcp-out:/data/out
volumes:
certs:
```
@@ -254,7 +287,7 @@ volumes:
git clone https://gitee.com/jinghui111/unicut.git
cd unicut
docker compose up -d unicut # 只启动编辑器
docker compose up -d # 全量(需 ../unicut-tts 源码)
docker compose up -d # 全量(含配音 unicut-tts 与 MCP unicut-mcp)
```
首次构建包含前端打包,约需几分钟。完成后浏览器访问:
@@ -266,6 +299,7 @@ http://服务器IP:3006
- 端口映射:`3006 → 80`(HTTP)、`3005 → 443`(HTTPS,自签证书,首次访问需接受一次证书警告);
- 首次进入按提示选择一个本地文件夹并授权,工程与素材即保存在该文件夹;
- 可选:在 `.env` 里配置 `HUB_KEY` / `HUB_PROJECT_ID` 接入授权中心(公告 / 版本检查),不配置不影响任何功能;
- 可选:在 `.env` 里配置 `MCP_LLM_KEY`(OpenAI 兼容网关 Key)启用 MCP 的文案生成,AI 客户端连接 `https://<站点>/mcp`,工具明细见 [mcp/README.md](./mcp/README.md);不配置则除大模型类工具外均可使用;
- 完整版另含配音服务 `unicut-tts`(本机 edge-tts),需单独部署该服务后用 `docker compose up -d` 全量启动;只用编辑器按上面的 `up -d unicut` 即可。
### 方式三:本地开发
+17
View File
@@ -23,5 +23,22 @@ services:
volumes:
- ../unicut-tts/audio:/data/audio
# MCP 服务(Model Context Protocol):AI 客户端经 https://<站点>/mcp 调用米剪能力
# (抓取正文配图 / 大模型口播文案 / edge-tts 配音 / 生成米剪工程 ZIP)。
# 密钥与站点地址从项目 .env 注入(MCP_LLM_KEY 等,勿提交仓库)。
unicut-mcp:
build: ./mcp
container_name: unicut-mcp
restart: always
environment:
- MCP_LLM_BASE=${MCP_LLM_BASE:-http://192.168.68.19:3099/v1}
- MCP_LLM_KEY=${MCP_LLM_KEY:-}
- MCP_LLM_MODEL=${MCP_LLM_MODEL:-mimo-v2.6-flash}
- MCP_PUBLIC_URL=${MCP_PUBLIC_URL:-https://127.0.0.1:3005}
- MCP_PROJECTS_DIR=${MCP_PROJECTS_DIR:-}
volumes:
# 生成的工程 ZIP / 配音 mp3 落盘(nginx /mcp-out/ 提供下载)
- ./mcp-out:/data/out
volumes:
certs:
+17
View File
@@ -0,0 +1,17 @@
# 米剪 MCP 服务:Streamable HTTP 端点 /mcp(8765),由 unicut nginx 同源反代
FROM docker.1ms.run/library/python:3.11-slim
ENV PYTHONUNBUFFERED=1 \
PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple \
PIP_DISABLE_PIP_VERSION_CHECK=1
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY server.py .
EXPOSE 8765
CMD ["python", "server.py"]
+107
View File
@@ -0,0 +1,107 @@
# 米剪 MCP 服务(unicut-mcp)
让 AI 客户端以 [Model Context Protocol](https://modelcontextprotocol.io) 工具的形式调用米剪的创作能力:抓文章、写口播、配语音、生成米剪工程,人在米剪里打开就能剪。
```
AI(Claude / Cursor / Hermes …)
│ MCP(Streamable HTTP)
▼
https://<米剪站点>/mcp ──► unicut-mcp 容器(本目录)
├─ grab_url 抓网页正文与配图(公众号文章可用)
├─ generate_script 大模型生成口播文案(OpenAI 兼容网关)
├─ tts / list_voices edge-tts 配音(unicut-tts 容器)
├─ create_video_project 生成米剪工程 ZIP → 米剪「工程文件」导入
└─ list/read/write_project 工作文件夹(文件夹模式)工程读写
```
## 工具一览
| 工具 | 作用 | 典型返回 |
| --- | --- | --- |
| `service_status` | 自检抓取/配音/大模型/工程目录可用性 | 各依赖状态 |
| `grab_url(url)` | 抓取网页正文与配图(规则与前端 `extractFromHtml` 一致:`#js_content` 优先、懒加载属性、12000 字截断) | `{title, text, images[]}` |
| `generate_script(source, requirement?, style?, target_seconds?, paragraph_count?)` | 大模型写口播文案,严格 JSON 输出 | `{title, paragraphs[], text}` |
| `list_voices()` | edge-tts 音色列表 | `{count, voices[]}` |
| `tts(text, voice?, rate?, pitch?, volume?)` | 合成配音,音频落盘并给出可播放 URL | `{url, path, duration}` |
| `create_video_project(title, images, script?, audio?, width?, height?, …)` | 下载素材 → 组装 MovieData → 打包工程 ZIP(`project.json + materials/`,与米剪导出格式同构,过 `checkNdata` 九键门禁) | `{zip_path, zip_url, duration, captions, elements}` |
| `list_projects()` / `read_project(id)` / `write_project(project)` | 工作文件夹模式下直接读写 `projects/*.json`(需配置 `MCP_PROJECTS_DIR`) | 工程 JSON |
`create_video_project` 细节:
- 图片校验与前端一致:≥160×120、≤10MB、识别失败自动跳过并在 `warnings` 里说明;
- 有配音时图片自动拉伸平铺铺满配音时长,无配音时每图 `image_duration`(默认 3 秒);
- `script` 按 `\n` 分段成字幕(单段超 24 字自动按句拆),按字数加权分配时间;
- 字幕/图片/音频元素结构照抄官方 demo 工程(`public/assets/demo/test-app.json`)真实字段。
## 连接方式
**端点**:`https://<米剪站点>/mcp`(Streamable HTTP;本机默认 `https://127.0.0.1:3005/mcp`,自签证书需客户端允许 insecure)。局域网内也可直连容器端口 `http://<主机>:8765/mcp`(默认只在 compose 网络内,未映射宿主端口)。
**Claude Desktop**(`claude_desktop_config.json`):
```json
{
"mcpServers": {
"unicut": {
"command": "npx",
"args": ["mcp-remote", "https://127.0.0.1:3005/mcp", "--allow-self-signed"]
}
}
}
```
**Claude Code / Cursor**:
```bash
claude mcp add unicut --transport http https://127.0.0.1:3005/mcp
# Cursor:~/.cursor/mcp.json → {"mcpServers":{"unicut":{"url":"https://127.0.0.1:3005/mcp"}}}
```
**Hermes**(`~/.hermes` 或项目 `config.yaml`):
```yaml
mcp:
servers:
unicut:
transport: http
url: https://127.0.0.1:3005/mcp
```
## 部署
compose 已内置 `unicut-mcp` 服务(本仓库 `docker-compose.yml`),密钥走项目 `.env`:
```bash
# .env(勿提交)
MCP_LLM_BASE=http://192.168.68.19:3099/v1 # OpenAI 兼容 /v1 网关
MCP_LLM_KEY=sk-xxxx # 网关 Key
MCP_LLM_MODEL=mimo-v2.6-flash
MCP_PUBLIC_URL=https://127.0.0.1:3005 # 对外站点地址(生成 zip_url / 音频 URL 用)
MCP_PROJECTS_DIR= # 可选:米剪「选择文件夹」的同一目录(文件夹模式)
```
```bash
docker compose up -d --build unicut-mcp
```
单独运行(不走 nginx):
```bash
docker build -t unicut-mcp ./mcp
docker run -d --name unicut-mcp --restart always \
-e MCP_LLM_KEY=sk-xxxx -e MCP_LLM_BASE=http://192.168.68.19:3099/v1 \
-p 8765:8765 -v "$PWD/mcp-out:/data/out" unicut-mcp
```
> 只有 `service_status` / `grab_url` / `generate_script` 需要网络依赖;`tts` 依赖同 compose 的 `unicut-tts`;不配置 `.env` 时除大模型类工具外均可使用。
## 验证
```bash
# 协议握手
curl -sk https://127.0.0.1:3005/mcp -H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'
```
产物下载:`https://<站点>/mcp-out/<工程名>.zip`;健康检查:经 nginx 用 `https://<站点>/mcp-health`,直连容器用 `http://<容器>:8765/health`(`/mcp/health` 不是端点,会被 SPA 回退成首页 HTML)。
+5
View File
@@ -0,0 +1,5 @@
mcp>=1.9.0,<2
httpx>=0.27.0
beautifulsoup4>=4.12.0
pillow>=10.0.0
mutagen>=1.47.0
+838
View File
@@ -0,0 +1,838 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
米剪(Unicut)MCP 服务
======================
让 AI 客户端(Claude Desktop / Claude Code / Cursor / Hermes 等)以 MCP 工具的形式
调用米剪的创作能力:
- grab_url 抓取网页正文与配图(公众号文章等,复用前端 extractFromHtml 规则)
- generate_script 大模型生成短视频口播文案
- tts / list_voices edge-tts 配音合成(unicut-tts)
- create_video_project 组装米剪工程 ZIP(project.json + materials/),在米剪「工程文件」一键导入
- list_projects / read_project 工作文件夹(文件夹模式)下的工程读写
- service_status 各依赖服务可用性自检
传输:Streamable HTTP,端点 /mcp(容器 8765,nginx 同源反代 https://<站点>/mcp)。
工程格式:与米剪 projectExport.ts 导出的 ZIP 完全同构——
project.json MovieData(title/poster/width/height/background/transitions/
captions/elements/resouces 九键齐全,过 checkNdata 门禁)
materials/{type}/{文件名} 素材相对路径
"""
from __future__ import annotations
import io
import ipaddress
import json
import os
import re
import socket
import time
import unicodedata
import uuid
import zipfile
from pathlib import Path
from urllib.parse import urljoin, urlparse
import httpx
from bs4 import BeautifulSoup
from mutagen.mp3 import MP3
from PIL import Image
from mcp.server.fastmcp import FastMCP
# ---------------------------------------------------------------- 环境配置
UNICUT_URL = os.getenv("UNICUT_URL", "http://unicut") # nginx 80(回环到 443)
TTS_URL = os.getenv("TTS_URL", "http://unicut-tts:8000")
LLM_BASE = os.getenv("MCP_LLM_BASE", "http://192.168.68.19:3099/v1").rstrip("/")
LLM_KEY = os.getenv("MCP_LLM_KEY", "")
LLM_MODEL = os.getenv("MCP_LLM_MODEL", "mimo-v2.6-flash")
PUBLIC_URL = os.getenv("MCP_PUBLIC_URL", "https://127.0.0.1:3005").rstrip("/")
PROJECTS_DIR = os.getenv("MCP_PROJECTS_DIR", "") # 文件夹模式工作目录(可空)
OUT_DIR = Path(os.getenv("MCP_OUT_DIR", "/data/out"))
IMG_MIN_W, IMG_MIN_H = 160, 120 # 与前端 sourceImport.ts 常量一致
IMG_MAX_BYTES = 10 * 1024 * 1024
HTTP_HEADERS = {
# 无 Referer:公众号图床 mmbiz.qpic.cn 带异源 Referer 时只给降级缩略图(实测)
"User-Agent": (
"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
"(KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36"
),
"Accept": "*/*",
"Accept-Language": "zh-CN,zh;q=0.9",
}
mcp = FastMCP(
"unicut",
instructions=(
"米剪(Unicut)在线视频剪辑器的 MCP 服务。"
"典型工作流:grab_url 抓文章 → generate_script 写口播 → tts 配音 → "
"create_video_project 生成米剪工程 ZIP → 用户在米剪「工程文件」导入即可出片。"
),
host="0.0.0.0",
port=8765,
)
def _client(timeout: float = 30.0) -> httpx.AsyncClient:
# trust_env=False:不读宿主代理变量,容器网络行为可预期
return httpx.AsyncClient(timeout=timeout, follow_redirects=True, trust_env=False)
# ---------------------------------------------------------------- 安全与小工具
def assert_public_url(url: str) -> str:
"""拒绝非 http(s) 与内网/回环目标(与 nginx /grab/ 的断言同规则,防 SSRF 打内网)"""
p = urlparse(url)
if p.scheme not in ("http", "https"):
raise ValueError(f"只支持 http/https URL:{url}")
host = p.hostname
if not host:
raise ValueError(f"URL 缺少主机名:{url}")
if host == "localhost" or host.endswith(".local"):
raise ValueError(f"拒绝内网/回环目标:{host}")
try:
infos = socket.getaddrinfo(host, None, proto=socket.IPPROTO_TCP)
except socket.gaierror as e:
raise ValueError(f"域名解析失败:{host}({e})")
for info in infos:
ip = ipaddress.ip_address(info[4][0])
if ip.is_private or ip.is_loopback or ip.is_link_local or ip.is_reserved or ip.is_multicast:
raise ValueError(f"拒绝内网地址目标:{host} → {ip}")
return str(p)
def _uid() -> str:
return uuid.uuid4().hex[:10]
def _slug(name: str) -> str:
"""文件名安全化:保留中英文数字,替换掉路径/控制字符"""
s = unicodedata.normalize("NFKC", name or "project")
s = re.sub(r'[\\/:*?"<>|\r\n\t]', "-", s).strip(" .")
return (s or "project")[:60]
def _ext_for(content_type: str, url: str, default: str = "bin") -> str:
ct = (content_type or "").split(";")[0].strip().lower()
table = {
"image/jpeg": "jpg", "image/jpg": "jpg", "image/png": "png",
"image/webp": "webp", "image/gif": "gif", "audio/mpeg": "mp3",
"audio/mp3": "mp3", "audio/wav": "wav", "audio/x-wav": "wav",
"audio/mp4": "m4a", "audio/aac": "aac",
}
if ct in table:
return table[ct]
suffix = Path(urlparse(url).path).suffix.lstrip(".").lower()
return suffix or default
def _is_local_path(s: str) -> bool:
return s.startswith("/") and not s.startswith("//")
# ---------------------------------------------------------------- 抓取(对齐前端 extractFromHtml)
_BLOCK_TAGS = [
"p", "h1", "h2", "h3", "h4", "h5", "h6", "li", "blockquote", "pre",
"td", "th", "figcaption", "tr", "div", "br",
]
_NOISE_TAGS = [
"script", "style", "noscript", "svg", "iframe", "canvas", "form",
"nav", "footer", "header", "aside", "button",
]
def extract_from_html(html: str, base_url: str) -> dict:
"""标题 / 正文 / 配图(相对转绝对、懒加载属性、去重)——与前端规则一致"""
soup = BeautifulSoup(html, "html.parser")
og = soup.find("meta", attrs={"property": "og:title"})
title = ((og.get("content") if og else None) or soup.title.get_text() if soup.title else "").strip()
for el in soup.find_all(_NOISE_TAGS):
el.decompose()
root = (
soup.find(id="js_content") # 公众号正文(微信页无 article/main)
or soup.find("article")
or soup.find("main")
or soup.find(attrs={"role": "main"})
or soup.body
or soup
)
# 图片:og:image 优先 + img 常见懒加载属性
seen: set[str] = set()
images: list[str] = []
def push(v: str | None) -> None:
if not v:
return
s = v.strip()
if not s or s.startswith("data:") or s.startswith("blob:"):
return
if re.search(r"\.svg(\?|#|$)", s, re.I):
return
try:
absu = urljoin(base_url, s)
except Exception:
return
if not re.match(r"^https?:", absu, re.I) or absu in seen:
return
seen.add(absu)
images.append(absu)
og_img = soup.find("meta", attrs={"property": "og:image"})
push(og_img.get("content") if og_img else None)
for img in root.find_all("img"):
for attr in ("src", "data-src", "data-original", "data-lazy-src", "data-actualsrc"):
push(img.get(attr))
# 正文:块级边界补换行 → textContent → 压空白(照抄前端清洗链)
for el in root.find_all(_BLOCK_TAGS):
el.insert(0, "\n")
el.append("\n")
text = (
root.get_text()
.replace("\r", "")
.replace("\t", " ")
.replace(" ", " ")
)
lines = [seg.strip() for seg in text.split("\n")]
kept: list[str] = []
for i, seg in enumerate(lines):
if seg or (i > 0 and lines[i - 1]):
kept.append(seg)
text = re.sub(r"\n{3,}", "\n\n", "\n".join(kept)).strip()[:12000]
return {"title": title, "text": text, "images": images[:20]}
async def fetch_bytes(url: str, *, allow_local_file: bool = True, timeout: float = 25.0) -> tuple[bytes, str]:
"""下载字节。url 可为公网 URL,或(默认允许)服务器本地文件路径。"""
if _is_local_path(url):
if not allow_local_file:
raise ValueError("此上下文不允许本地路径")
p = Path(url)
if not p.is_file():
raise ValueError(f"本地文件不存在:{url}")
return p.read_bytes(), ""
assert_public_url(url)
async with _client(timeout) as c:
r = await c.get(url, headers=HTTP_HEADERS)
r.raise_for_status()
return r.content, r.headers.get("content-type", "")
# ---------------------------------------------------------------- 大模型
async def _llm_chat(messages: list[dict], *, temperature: float = 0.7, max_tokens: int = 8000) -> str:
if not LLM_KEY:
raise ValueError("未配置 MCP_LLM_KEY(compose .env),无法调用大模型")
async with _client(timeout=180.0) as c:
r = await c.post(
f"{LLM_BASE}/chat/completions",
headers={"Authorization": f"Bearer {LLM_KEY}"},
json={
"model": LLM_MODEL,
"messages": messages,
"temperature": temperature,
"max_tokens": max_tokens,
},
)
if r.status_code >= 400:
raise ValueError(f"大模型网关 {r.status_code}: {r.text[:300]}")
data = r.json()
return data["choices"][0]["message"]["content"]
def _parse_json_block(s: str) -> dict | None:
s = s.strip()
m = re.search(r"\{.*\}", s, re.S)
if not m:
return None
try:
return json.loads(m.group(0))
except json.JSONDecodeError:
# 容忍 ```json 围栏 / 尾逗号
cleaned = re.sub(r"^```(?:json)?|```$", "", m.group(0), flags=re.M).strip()
cleaned = re.sub(r",\s*([}\]])", r"\1", cleaned)
try:
return json.loads(cleaned)
except json.JSONDecodeError:
return None
# ---------------------------------------------------------------- 工程组装(MovieData)
def _split_captions(script: str, total: float, audio_dur: float | None) -> list[dict]:
"""口播文案 → 字幕段(按字符数加权切分时间轴;有配音时铺满配音时长)"""
raw = [ln.strip() for ln in re.split(r"\n+", script) if ln.strip()]
lines: list[str] = []
for ln in raw:
if len(ln) <= 24:
lines.append(ln)
else: # 长段按句切(保留标点)
parts = [s for s in re.split(r"(?<=[。!?!?;;])", ln) if s.strip()]
lines.extend([p.strip() for p in parts] if parts else [ln])
if not lines:
return []
timeline = audio_dur if audio_dur else total
if timeline <= 0:
return []
weights = [max(len(x), 1) for x in lines]
s = sum(weights)
caps: list[dict] = []
t = 0.0
for i, (line, w) in enumerate(zip(lines, weights)):
dur = timeline * w / s
if i == len(lines) - 1:
dur = max(timeline - t, 0.3) # 末段补到结尾,避免浮点欠尾
caps.append(_caption_element(line, t, max(dur, 0.3)))
t += dur
return caps
def _caption_element(text: str, start: float, dur: float, height: int = 1920, fill: str = "rgba(255, 255, 255, 1)") -> dict:
"""字幕元素:结构照抄 demo/test-app.json 的真实字幕(字号缩放随画布高度)"""
scale = round(height / 1920, 4) or 1
return {
"id": _uid(),
"name": text,
"text": text,
"textStyle": {
"radius": [0, 0, 0, 0],
"type": "newTextStyle",
"direction": "horizontal",
"fillType": "normal",
"fill": fill,
"padding": [5, 5, 5, 5],
"fontSize": 200,
"letterSpacing": 0,
"lineHeight": 200,
"align": "center",
"fontFamily": "抖音美好体",
"stroke": {"color": "rgba(51, 51, 51, 1)", "lineJoin": "round", "lineWidth": 18},
"fontFamilyURL": "/assets/fonts/抖音美好体.otf",
},
"style": {
"x": 540,
"y": round(height * 0.77, 2),
"alpha": 1,
"width": 392.96,
"height": 67.2,
},
"type": "caption",
"trackIndex": 0.5,
"startTime": round(start, 3),
"duration": round(dur, 3),
"textScale": round(0.32 * scale, 4),
"scale": 1,
"templateEnable": True,
"sceneIndex": 1,
}
def _image_element(resource_id: str, name: str, start: float, dur: float, w: int, h: int, idx: int) -> dict:
"""图片元素:字段集对齐 demo 视频元素(同一 BaseElement 血统)+ ImageElement 特有 matting"""
return {
"id": _uid(),
"name": name,
"type": "image",
"trackIndex": 0,
"resourceId": resource_id,
"startTime": round(start, 3),
"duration": round(dur, 3),
"style": {"width": w, "height": h, "x": w / 2, "y": h / 2}, # x/y 是中心点(demo 实证)
"blendMode": 0,
"flipx": False,
"scale": 1,
"filters": [],
"matting": {
"enabled": False,
"color": "#00FF00",
"similarityThreshold": 0.5,
"emergence": 0.5,
},
"sceneIndex": idx + 1,
"templateEnable": False,
}
def _audio_element(resource_id: str, dur: float, volume: float, name: str = "配音.mp3") -> dict:
return {
"id": _uid(),
"name": name,
"type": "audio",
"trackIndex": 1,
"resourceId": resource_id,
"startTime": 0,
"duration": round(dur, 3),
"muted": False,
"volume": volume,
"speed": 1,
"curveSpeed": False,
"clipTime": 0,
"fadeInTime": 0,
"fadeOutTime": 0,
}
def build_movie_data(
title: str,
w: int,
h: int,
images: list[dict], # [{res_id, zip_rel, name, iw, ih, mime, size}]
audio: dict | None, # {res_id, zip_rel, name, dur, mime, size}
caption_lines: str,
with_caption: bool,
caption_fill: str,
voice_volume: float,
image_duration: float,
) -> dict:
now = int(time.time() * 1000)
resouces: list[dict] = []
elements: list[dict] = []
n = len(images)
if audio:
total = float(audio["dur"])
span = total / n if n else 0.0
else:
span = float(image_duration)
total = n * span
for i, im in enumerate(images):
resouces.append({
"id": im["res_id"],
"name": im["name"],
"type": "image",
"url": im["zip_rel"],
"originId": im["res_id"],
"thumb": "",
"styleSize": {"width": im["iw"], "height": im["ih"]},
"fileType": im["mime"],
"from": "user",
"mustFetch": False,
"attrs": {"width": im["iw"], "height": im["ih"], "size": im["size"], "fileType": im["mime"]},
})
elements.append(_image_element(im["res_id"], im["name"], i * span, span, w, h, i))
if audio:
resouces.append({
"id": audio["res_id"],
"name": audio["name"],
"type": "audio",
"url": audio["zip_rel"],
"originId": audio["res_id"],
"thumb": "",
"styleSize": {"width": 0, "height": 0},
"fileType": audio["mime"],
"from": "user",
"mustFetch": False,
"duration": audio["dur"],
"attrs": {"size": audio["size"], "fileType": audio["mime"], "duration": audio["dur"]},
})
elements.append(_audio_element(audio["res_id"], audio["dur"], voice_volume, audio["name"]))
captions = _split_captions(caption_lines, total, audio["dur"] if audio else None) if with_caption else []
# checkNdata 门禁九键 + demo 同款元数据
return {
"title": title,
"createTime": now,
"updateTime": now,
"poster": "",
"version": "1.1.0",
"width": w,
"height": h,
"duration": round(total, 3),
"background": {},
"_hideLock": {},
"transitions": [],
"cameras": [],
"captions": captions,
"elements": elements,
"resouces": resouces,
}
def write_project_zip(movie: dict, files: dict[str, bytes], zip_path: Path) -> None:
"""project.json + materials/{type}/{file} —— 与 projectExport.ts 导出同构"""
zip_path.parent.mkdir(parents=True, exist_ok=True)
with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as z:
z.writestr("project.json", json.dumps(movie, ensure_ascii=False))
for rel, blob in files.items():
z.writestr(rel, blob)
# ---------------------------------------------------------------- MCP 工具
@mcp.tool()
async def service_status() -> dict:
"""自检:抓取反代 / 配音服务 / 大模型网关 / 工程目录 的可用性"""
out: dict[str, object] = {"time": time.strftime("%Y-%m-%d %H:%M:%S")}
try:
async with _client(8) as c:
r = await c.get(f"{UNICUT_URL}/")
out["unicut_nginx"] = f"OK({r.status_code})"
except Exception as e:
out["unicut_nginx"] = f"FAIL: {e}"
try:
async with _client(8) as c:
r = await c.get(f"{TTS_URL}/voices")
v = r.json()
n = len(v) if isinstance(v, list) else len(v.get("voices", [])) if isinstance(v, dict) else 0
out["tts"] = f"OK, {n} 个音色"
except Exception as e:
out["tts"] = f"FAIL: {e}"
out["llm"] = {
"base": LLM_BASE,
"model": LLM_MODEL,
"key": f"已配置({len(LLM_KEY)}位)" if LLM_KEY else "未配置 MCP_LLM_KEY",
}
out["projects_dir"] = PROJECTS_DIR or "未配置(文件夹模式工具不可用)"
out["out_dir"] = str(OUT_DIR)
out["public_url"] = PUBLIC_URL
return out
@mcp.tool()
async def grab_url(url: str, max_images: int = 20) -> dict:
"""抓取网页正文与配图(支持公众号 mp.weixin.qq.com 文章)。
返回 {ok, url, title, text(≤12000字), images[](绝对地址, 最多 max_images 张)}。
图片只是候选地址;create_video_project 下载时会再校验尺寸(≥160x120)与大小(≤10MB)。
"""
assert_public_url(url)
async with _client(30) as c:
r = await c.get(url, headers=HTTP_HEADERS)
r.raise_for_status()
final_url = str(r.url)
ctype = r.headers.get("content-type", "")
if "html" not in ctype and "xml" not in ctype:
# 非 HTML(如直接是文本/JSON):原样返回文本
return {"ok": True, "url": final_url, "title": final_url, "text": r.text[:12000], "images": []}
ext = extract_from_html(r.text, final_url)
return {
"ok": True,
"url": final_url,
"title": ext["title"],
"text": ext["text"],
"images": ext["images"][: max(0, min(int(max_images), 20))],
}
@mcp.tool()
async def generate_script(
source: str,
requirement: str = "",
style: str = "口播解说",
target_seconds: int = 60,
paragraph_count: int = 6,
) -> dict:
"""大模型生成短视频口播文案。
source:素材文本(抓取的文章正文)或主题描述;requirement:额外要求(受众/卖点/关键词)。
返回 {ok, title, paragraphs[], text};paragraphs 每段一句,直接对应字幕与时间轴分段。
"""
sys_prompt = (
"你是短视频口播文案编剧。只输出一个 JSON 对象,不要任何解释、不要 markdown 围栏:\n"
'{"title": "视频标题", "paragraphs": ["第一段口播", "第二段口播", ...]}\n'
"要求:口语化、有钩子、每段 1~2 句话适合做字幕;总字数 ≈ 目标秒数 × 4.5 字。"
)
user_prompt = (
f"风格:{style}\n目标时长:约 {target_seconds} 秒\n段落数:约 {paragraph_count} 段\n"
)
if requirement:
user_prompt += f"补充要求:{requirement}\n"
user_prompt += f"\n素材:\n{source[:8000]}"
content = await _llm_chat([
{"role": "system", "content": sys_prompt},
{"role": "user", "content": user_prompt},
])
data = _parse_json_block(content)
if data and isinstance(data.get("paragraphs"), list):
paras = [str(p).strip() for p in data["paragraphs"] if str(p).strip()]
title = str(data.get("title") or "").strip()
else: # 兜底:按行切
paras = [ln.strip() for ln in content.splitlines() if ln.strip()]
title = ""
if not paras:
raise ValueError(f"模型未返回可用文案:{content[:300]}")
return {"ok": True, "title": title, "paragraphs": paras, "text": "\n".join(paras)}
@mcp.tool()
async def list_voices() -> dict:
"""列出配音(edge-tts)可用音色 → {ok, count, voices[]}"""
async with _client(15) as c:
r = await c.get(f"{TTS_URL}/voices")
r.raise_for_status()
v = r.json()
voices = v if isinstance(v, list) else v.get("voices", [])
return {"ok": True, "count": len(voices), "voices": voices}
@mcp.tool()
async def tts(
text: str,
voice: str = "zh-CN-XiaoxiaoNeural",
rate: str = "+0%",
pitch: str = "+0Hz",
volume: str = "+0%",
) -> dict:
"""edge-tts 配音合成。返回 {ok, url(站点可直接播放), path(服务器文件), duration(秒), voice}。
text 建议 ≤2000 字;传给 create_video_project 的 audio 用 path。
"""
if not text.strip():
raise ValueError("text 不能为空")
async with _client(180) as c:
r = await c.post(
f"{TTS_URL}/tts",
json={"text": text, "voice": voice, "rate": rate, "pitch": pitch, "volume": volume},
)
if r.status_code >= 400:
raise ValueError(f"TTS {r.status_code}: {r.text[:300]}")
j = r.json()
if not isinstance(j, dict) or not j.get("url"):
raise ValueError(f"TTS 返回异常:{str(j)[:300]}")
u = str(j["url"])
internal = TTS_URL + (u[len("/localtts"):] if u.startswith("/localtts") else u)
ar = await c.get(internal)
ar.raise_for_status()
blob = ar.content
if not blob:
raise ValueError("TTS 音频为空")
audio_dir = OUT_DIR / "audio"
audio_dir.mkdir(parents=True, exist_ok=True)
name = Path(urlparse(u).path).name or f"{_uid()}.mp3"
dest = audio_dir / name
dest.write_bytes(blob)
try:
dur = float(MP3(io.BytesIO(blob)).info.length)
except Exception:
dur = 0.0
public = PUBLIC_URL + (u if u.startswith("/") else "/" + u)
return {"ok": True, "url": public, "path": str(dest), "duration": round(dur, 3), "voice": voice}
@mcp.tool()
async def create_video_project(
title: str,
images: list[str] | None = None,
script: str = "",
audio: str = "",
width: int = 1080,
height: int = 1920,
image_duration: float = 3.0,
caption: bool = True,
caption_fill: str = "rgba(255, 255, 255, 1)",
voice_volume: float = 1.0,
) -> dict:
"""生成米剪工程 ZIP(project.json + materials/),在米剪顶部「工程文件」导入即可出片。
images:图片 URL 或服务器本地路径(tts 工具产出的 path 也行);script:口播文案→自动做字幕
(按\\n分段,单段过长自动按句拆);audio:配音 URL/path(通常来自 tts 工具)。
有配音时图片自动拉伸铺满配音时长,否则每图 image_duration 秒。
返回 {ok, zip_path, zip_url(可直接下载), id, duration, images_used, captions, elements}。
"""
images = images or []
warnings: list[str] = []
# ---- 图片收集与校验
im_items: list[dict] = []
files: dict[str, bytes] = {}
for raw in images:
if len(im_items) >= 30:
warnings.append("图片超过 30 张,多余已丢弃")
break
try:
blob, ctype = await fetch_bytes(raw)
except Exception as e:
warnings.append(f"图片下载失败(跳过){raw[:80]}: {e}")
continue
if len(blob) > IMG_MAX_BYTES:
warnings.append(f"图片超 10MB(跳过){raw[:80]}")
continue
try:
im = Image.open(io.BytesIO(blob))
iw, ih = im.size
except Exception as e:
warnings.append(f"图片无法识别(跳过){raw[:80]}: {e}")
continue
if iw < IMG_MIN_W or ih < IMG_MIN_H:
warnings.append(f"图片太小 {iw}x{ih}(跳过){raw[:80]}")
continue
ext = _ext_for(ctype, raw, "jpg")
if ext not in ("jpg", "png", "webp", "gif"):
ext = "jpg" if im.format in ("JPEG", "PNG", "WEBP", "GIF") else "jpg"
fname = f"{uuid.uuid4().hex}.{ext}"
rel = f"materials/image/{fname}"
files[rel] = blob
im_items.append({
"res_id": _uid(),
"name": Path(urlparse(raw).path).name or fname,
"zip_rel": rel,
"iw": iw, "ih": ih,
"mime": f"image/{'jpeg' if ext == 'jpg' else ext}",
"size": len(blob),
})
# ---- 配音收集
aud: dict | None = None
if audio:
try:
blob, ctype = await fetch_bytes(audio)
ext = _ext_for(ctype, audio, "mp3")
fname = f"voice_{uuid.uuid4().hex[:8]}.{ext}"
rel = f"materials/audio/{fname}"
files[rel] = blob
try:
dur = float(MP3(io.BytesIO(blob)).info.length)
except Exception:
dur = 0.0
if dur <= 0:
warnings.append("配音时长无法识别,时间轴将按图片时长排布")
aud = None
else:
aud = {
"res_id": _uid(),
"name": Path(urlparse(audio).path).name or fname,
"zip_rel": rel,
"dur": dur,
"mime": f"audio/{'mpeg' if ext == 'mp3' else ext}",
"size": len(blob),
}
except Exception as e:
warnings.append(f"配音获取失败(忽略): {e}")
if not im_items and not aud:
raise ValueError("至少需要 1 张可用图片或 1 段配音")
# ---- 组装 MovieData
movie = build_movie_data(
title=title, w=int(width), h=int(height),
images=im_items, audio=aud,
caption_lines=script, with_caption=bool(caption and script.strip()),
caption_fill=caption_fill, voice_volume=float(voice_volume),
image_duration=float(image_duration),
)
# ---- 落盘 ZIP
ts = time.strftime("%Y%m%d-%H%M%S")
zip_name = f"{_slug(title)}-{ts}.zip"
zip_path = OUT_DIR / zip_name
write_project_zip(movie, files, zip_path)
# ---- 门禁自检:checkNdata 九键
required = ["title", "poster", "width", "height", "background", "transitions", "captions", "elements", "resouces"]
missing = [k for k in required if k not in movie]
if missing:
raise ValueError(f"工程缺少必需键:{missing}")
return {
"ok": True,
"id": f"mcp-{ts}-{_uid()[:6]}",
"zip_path": str(zip_path),
"zip_url": f"{PUBLIC_URL}/mcp-out/{zip_name}",
"duration": movie["duration"],
"size": zip_path.stat().st_size,
"images_used": len(im_items),
"has_audio": bool(aud),
"captions": len(movie["captions"]),
"elements": len(movie["elements"]),
"warnings": warnings,
"import_hint": "下载该 ZIP → 米剪顶部「工程文件」→ 导入(需先选择过工作文件夹)",
}
@mcp.tool()
async def list_projects() -> dict:
"""列出工作文件夹(文件夹模式 PROJECTS_DIR)里的米剪工程 → {ok, dir, projects[]}"""
if not PROJECTS_DIR:
return {"ok": False, "error": "未配置 MCP_PROJECTS_DIR(compose .env)。文件夹模式需把它指到米剪里选择的同一目录"}
pdir = Path(PROJECTS_DIR) / "projects"
if not pdir.is_dir():
return {"ok": True, "dir": str(pdir), "projects": []}
items = []
for f in sorted(pdir.glob("*.json")):
try:
d = json.loads(f.read_text(encoding="utf-8"))
items.append({
"file": f.name,
"id": d.get("id") or f.stem,
"title": d.get("title") or f.stem,
"width": d.get("width"),
"height": d.get("height"),
"updatedAt": d.get("updatedAt"),
})
except Exception as e:
items.append({"file": f.name, "error": str(e)})
return {"ok": True, "dir": str(pdir), "projects": items}
@mcp.tool()
async def read_project(project_id: str) -> dict:
"""读取工作文件夹里的工程 JSON(folder 形态 {id,title,width,height,data})"""
if not PROJECTS_DIR:
return {"ok": False, "error": "未配置 MCP_PROJECTS_DIR"}
f = Path(PROJECTS_DIR) / "projects" / f"{_slug(project_id)}.json"
if not f.is_file():
return {"ok": False, "error": f"工程不存在:{f}"}
data = json.loads(f.read_text(encoding="utf-8"))
return {"ok": True, "path": str(f), "project": data}
@mcp.tool()
async def write_project(project: dict) -> dict:
"""写入/更新工作文件夹工程(folder 形态:{id,title,width,height,data})。谨慎调用:会覆盖同 id 文件"""
if not PROJECTS_DIR:
return {"ok": False, "error": "未配置 MCP_PROJECTS_DIR"}
if not isinstance(project, dict) or not project.get("data"):
return {"ok": False, "error": "需要 {id,title,width,height,data(MovieData)} 结构"}
pid = _slug(str(project.get("id") or _uid()))
project["id"] = pid
project["updatedAt"] = int(time.time() * 1000)
pdir = Path(PROJECTS_DIR) / "projects"
pdir.mkdir(parents=True, exist_ok=True)
f = pdir / f"{pid}.json"
f.write_text(json.dumps(project, ensure_ascii=False), encoding="utf-8")
return {"ok": True, "path": str(f), "id": pid}
# ---------------------------------------------------------------- 静态产物与健康检查路由
try:
from starlette.responses import FileResponse, JSONResponse
@mcp.custom_route("/health", methods=["GET"])
async def _health(_req):
return JSONResponse({"ok": True, "service": "unicut-mcp"})
@mcp.custom_route("/out/{name}", methods=["GET"])
async def _out(req):
name = Path(req.path_params["name"]).name # 防目录穿越
f = OUT_DIR / name
if not f.is_file():
return JSONResponse({"ok": False, "error": "not found"}, status_code=404)
return FileResponse(f, filename=name)
except AttributeError: # 老版本 SDK 无 custom_route,静默降级
pass
if __name__ == "__main__":
OUT_DIR.mkdir(parents=True, exist_ok=True)
mcp.run(transport="streamable-http")
+43
View File
@@ -116,6 +116,9 @@ server {
proxy_set_header Host $proxy_host;
proxy_set_header Connection "";
proxy_set_header X-Grab-Url "";
# 公众号图片防盗链:带异源 Referer 时 mmbiz.qpic.cn 只给降级缩略图
# (实测 2KB/140x140,无 Referer 则 47KB/原尺寸),向上游清空该头
proxy_set_header Referer "";
proxy_http_version 1.1;
proxy_read_timeout 20s;
proxy_send_timeout 20s;
@@ -143,6 +146,46 @@ server {
client_max_body_size 8m;
}
# MCP 服务(Model Context Protocol,unicut-mcp 容器)同源反代:
# AI 客户端(Claude Desktop / Claude Code / Cursor / Hermes 等)用
# https://<站点>/mcp
# 连接即可把米剪当工具用(抓取文章 / 生成口播文案 / edge-tts 配音 / 生成工程 ZIP)。
# Streamable HTTP 走 SSE 流:必须关缓冲、保持 HTTP/1.1 长连接、拉长读超时。
# 注:/mcp-out 前缀比 /mcp 更长,nginx 最长前缀匹配规则会优先后者,不冲突。
location /mcp {
proxy_pass http://unicut-mcp:8765;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Connection "";
proxy_buffering off;
proxy_request_buffering off;
proxy_cache off;
proxy_read_timeout 600s;
proxy_send_timeout 600s;
client_max_body_size 20m;
}
# MCP 生成产物下载(工程 ZIP、配音 mp3),配合米剪「工程文件」导入
location /mcp-out/ {
proxy_pass http://unicut-mcp:8765/out/;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_read_timeout 300s;
client_max_body_size 100m;
}
# MCP 健康检查(容器 /health;注意 /mcp/health 不是端点,SPA 会回退成首页 HTML)
location = /mcp-health {
proxy_pass http://unicut-mcp:8765/health;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_read_timeout 10s;
}
# UNICUT 在线服务反代(账号/素材/AI 等在线能力), 对应官方 dev server 的代理配置
location /api/ {
proxy_pass https://video.h5ds.com;
+1 -1
View File
@@ -1,7 +1,7 @@
{
"name": "aacut",
"private": true,
"version": "1.0.0",
"version": "1.0.1",
"scripts": {
"dev": "vite",
"build": "yarn clean && node scripts/patch-sdk-offline.mjs && yarn build:worker && vite build",
+1
View File
@@ -80,6 +80,7 @@ export const enUS = {
my_upload: 'Upload',
my_upload_mobile_link: 'Upload Mobile',
my_upload_refresh: 'Refresh List',
my_download_more: 'Download More Assets',
my_tab_material: 'Current Materials',
my_tab_all_material: 'All Materials',
my_tab_no_material: 'No materials in this project. Please upload first.',
+1
View File
@@ -91,6 +91,7 @@ export const zhCN = {
my_upload: '上传素材',
my_upload_mobile_link: '手机上传',
my_upload_refresh: '刷新列表',
my_download_more: '下载更多素材',
my_tab_material: '当前素材',
my_tab_all_material: '全部素材',
my_tab_no_material: '该项目暂无素材,请先上传',
@@ -12,7 +12,7 @@ import {
Pagination,
Empty,
} from '@douyinfe/semi-ui';
import { Voice, Left, VoiceInput, Upload as UploadIcon, DeleteFive, Close, Refresh } from '@icon-park/react';
import { Voice, Left, VoiceInput, Upload as UploadIcon, DeleteFive, Close, Refresh, Download } from '@icon-park/react';
import { useCallback, useEffect, useReducer, useRef, useState } from 'react';
import { getUploadBeforeData, isAPNG } from '@pages/editor/tools/uploadBeforeData';
import { pubsub, util } from '@utils/index';
@@ -497,6 +497,14 @@ export default function List(props: IProps) {
setShowType('tts');
},
},
{
node: 'item',
icon: <Download theme="outline" size="20" fill="var(--theme-icon)" />,
name: language.val('my_download_more'),
onClick: () => {
window.open('https://doc.88531.cn/docs/mrfx/mrfx-1g99qum57n5qk', '_blank');
},
},
]}
trigger="click"
position="bottomRight"
+3
View File
@@ -313,6 +313,9 @@ export function extractFromHtml(
.forEach(el => el.remove());
const root =
// 公众号文章正文在 #js_content(页面其余部分是账号栏/留言/分享等噪音),
// 必须优先于 article/main 选择器——微信页两者都没有,否则会退化到整页 body
dom.querySelector('#js_content') ||
dom.querySelector('article') ||
dom.querySelector('main') ||
dom.querySelector('[role="main"]') ||
+1 -1
View File
@@ -4929,7 +4929,7 @@ var drawChart = (function (exports) {
</script>
<script>
/*<!--*/
const data = {"version":2,"tree":{"name":"root","children":[{"name":"homeAssets/js/opencv-ts-26caf77c.js","children":[{"uid":"21e0ef06-1","name":"\u0000commonjsHelpers.js"},{"uid":"21e0ef06-3","name":"__vite-browser-external"},{"uid":"21e0ef06-5","name":"\u0000__vite-browser-external?commonjs-proxy"},{"name":"\u0000C:/E/unicut/node_modules/opencv-ts/src/opencv.js?commonjs-module","uid":"21e0ef06-7"},{"name":"C:/E/unicut/node_modules/opencv-ts/src","children":[{"uid":"21e0ef06-9","name":"opencv.js"},{"uid":"21e0ef06-11","name":"opencv.js?commonjs-entry"}]}]},{"name":"homeAssets/js/main-c4b3b641.js","children":[{"name":"\u0000vite","children":[{"uid":"21e0ef06-13","name":"modulepreload-polyfill"},{"uid":"21e0ef06-2639","name":"preload-helper"}]},{"name":"\u0000C:/E/unicut/node_modules","children":[{"name":"react","children":[{"uid":"21e0ef06-15","name":"jsx-runtime.js?commonjs-module"},{"name":"cjs","children":[{"uid":"21e0ef06-17","name":"react-jsx-runtime.production.min.js?commonjs-exports"},{"uid":"21e0ef06-21","name":"react.production.min.js?commonjs-exports"}]},{"uid":"21e0ef06-19","name":"index.js?commonjs-module"}]},{"name":"react-dom","children":[{"uid":"21e0ef06-31","name":"index.js?commonjs-module"},{"name":"cjs/react-dom.production.min.js?commonjs-exports","uid":"21e0ef06-33"},{"uid":"21e0ef06-47","name":"client.js?commonjs-exports"}]},{"name":"scheduler","children":[{"uid":"21e0ef06-35","name":"index.js?commonjs-module"},{"name":"cjs/scheduler.production.min.js?commonjs-exports","uid":"21e0ef06-37"}]},{"name":"axios/lib/axios.js?commonjs-module","uid":"21e0ef06-57"},{"name":"pubsub-js/src/pubsub.js?commonjs-module","uid":"21e0ef06-123"},{"name":"dayjs/dayjs.min.js?commonjs-module","uid":"21e0ef06-127"},{"name":"simple-query-string/src/simplequerystring.js?commonjs-module","uid":"21e0ef06-131"},{"name":"dexie/dist/dexie.min.js?commonjs-module","uid":"21e0ef06-135"},{"name":"ignore/index.js?commonjs-module","uid":"21e0ef06-143"},{"name":"classnames/index.js?commonjs-module","uid":"21e0ef06-277"},{"name":"prop-types/index.js?commonjs-module","uid":"21e0ef06-281"},{"name":"lodash","children":[{"uid":"21e0ef06-325","name":"isBuffer.js?commonjs-module"},{"uid":"21e0ef06-339","name":"_nodeUtil.js?commonjs-module"},{"uid":"21e0ef06-369","name":"_cloneBuffer.js?commonjs-module"}]},{"name":"copy-text-to-clipboard/index.js?commonjs-module","uid":"21e0ef06-599"},{"name":"date-fns-tz","children":[{"name":"format","children":[{"uid":"21e0ef06-1309","name":"index.js?commonjs-module"},{"name":"formatters/index.js?commonjs-module","uid":"21e0ef06-1459"}]},{"name":"_lib","children":[{"name":"tzIntlTimeZoneName/index.js?commonjs-module","uid":"21e0ef06-1461"},{"name":"tzParseTimezone/index.js?commonjs-module","uid":"21e0ef06-1465"},{"name":"tzTokenizeDate/index.js?commonjs-module","uid":"21e0ef06-1467"},{"name":"newDateUTC/index.js?commonjs-module","uid":"21e0ef06-1471"},{"name":"tzPattern/index.js?commonjs-module","uid":"21e0ef06-1481"}]},{"name":"toDate/index.js?commonjs-module","uid":"21e0ef06-1479"},{"name":"formatInTimeZone/index.js?commonjs-module","uid":"21e0ef06-1489"},{"name":"utcToZonedTime/index.js?commonjs-module","uid":"21e0ef06-1499"},{"name":"getTimezoneOffset/index.js?commonjs-module","uid":"21e0ef06-1505"},{"name":"zonedTimeToUtc/index.js?commonjs-module","uid":"21e0ef06-1509"}]},{"name":"date-fns","children":[{"name":"format/index.js?commonjs-module","uid":"21e0ef06-1311"},{"name":"isValid/index.js?commonjs-module","uid":"21e0ef06-1317"},{"name":"isDate/index.js?commonjs-module","uid":"21e0ef06-1319"},{"name":"_lib","children":[{"name":"requiredArgs/index.js?commonjs-module","uid":"21e0ef06-1325"},{"name":"toInteger/index.js?commonjs-module","uid":"21e0ef06-1341"},{"name":"format","children":[{"name":"formatters/index.js?commonjs-module","uid":"21e0ef06-1349"},{"name":"lightFormatters/index.js?commonjs-module","uid":"21e0ef06-1395"},{"name":"longFormatters/index.js?commonjs-module","uid":"21e0ef06-1401"}]},{"name":"getUTCDayOfYear/index.js?commonjs-module","uid":"21e0ef06-1351"},{"name":"getUTCISOWeek/index.js?commonjs-module","uid":"21e0ef06-1355"},{"name":"startOfUTCISOWeek/index.js?commonjs-module","uid":"21e0ef06-1357"},{"name":"startOfUTCISOWeekYear/index.js?commonjs-module","uid":"21e0ef06-1361"},{"name":"getUTCISOWeekYear/index.js?commonjs-module","uid":"21e0ef06-1363"},{"name":"getUTCWeek/index.js?commonjs-module","uid":"21e0ef06-1371"},{"name":"startOfUTCWeek/index.js?commonjs-module","uid":"21e0ef06-1373"},{"name":"defaultOptions/index.js?commonjs-exports","uid":"21e0ef06-1375"},{"name":"startOfUTCWeekYear/index.js?commonjs-module","uid":"21e0ef06-1381"},{"name":"getUTCWeekYear/index.js?commonjs-module","uid":"21e0ef06-1383"},{"name":"addLeadingZeros/index.js?commonjs-module","uid":"21e0ef06-1391"},{"name":"getTimezoneOffsetInMilliseconds/index.js?commonjs-module","uid":"21e0ef06-1405"},{"name":"protectedTokens/index.js?commonjs-exports","uid":"21e0ef06-1409"},{"name":"defaultLocale/Line truncated
const data = {"version":2,"tree":{"name":"root","children":[{"name":"homeAssets/js/opencv-ts-26caf77c.js","children":[{"uid":"ec5f2780-1","name":"\u0000commonjsHelpers.js"},{"uid":"ec5f2780-3","name":"__vite-browser-external"},{"uid":"ec5f2780-5","name":"\u0000__vite-browser-external?commonjs-proxy"},{"name":"\u0000/node_modules/opencv-ts/src/opencv.js?commonjs-module","uid":"ec5f2780-7"},{"name":"node_modules/opencv-ts/src","children":[{"uid":"ec5f2780-9","name":"opencv.js"},{"uid":"ec5f2780-11","name":"opencv.js?commonjs-entry"}]}]},{"name":"homeAssets/js/main-b9519a96.js","children":[{"name":"\u0000vite","children":[{"uid":"ec5f2780-13","name":"modulepreload-polyfill"},{"uid":"ec5f2780-2639","name":"preload-helper"}]},{"name":"\u0000/node_modules","children":[{"name":"react","children":[{"uid":"ec5f2780-15","name":"jsx-runtime.js?commonjs-module"},{"name":"cjs","children":[{"uid":"ec5f2780-17","name":"react-jsx-runtime.production.min.js?commonjs-exports"},{"uid":"ec5f2780-21","name":"react.production.min.js?commonjs-exports"}]},{"uid":"ec5f2780-19","name":"index.js?commonjs-module"}]},{"name":"react-dom","children":[{"uid":"ec5f2780-33","name":"index.js?commonjs-module"},{"name":"cjs/react-dom.production.min.js?commonjs-exports","uid":"ec5f2780-35"},{"uid":"ec5f2780-49","name":"client.js?commonjs-exports"}]},{"name":"scheduler","children":[{"uid":"ec5f2780-37","name":"index.js?commonjs-module"},{"name":"cjs/scheduler.production.min.js?commonjs-exports","uid":"ec5f2780-39"}]},{"name":"axios/lib/axios.js?commonjs-module","uid":"ec5f2780-59"},{"name":"pubsub-js/src/pubsub.js?commonjs-module","uid":"ec5f2780-125"},{"name":"dayjs/dayjs.min.js?commonjs-module","uid":"ec5f2780-129"},{"name":"simple-query-string/src/simplequerystring.js?commonjs-module","uid":"ec5f2780-133"},{"name":"dexie/dist/dexie.min.js?commonjs-module","uid":"ec5f2780-137"},{"name":"ignore/index.js?commonjs-module","uid":"ec5f2780-145"},{"name":"classnames/index.js?commonjs-module","uid":"ec5f2780-279"},{"name":"prop-types/index.js?commonjs-module","uid":"ec5f2780-283"},{"name":"lodash","children":[{"uid":"ec5f2780-327","name":"isBuffer.js?commonjs-module"},{"uid":"ec5f2780-341","name":"_nodeUtil.js?commonjs-module"},{"uid":"ec5f2780-371","name":"_cloneBuffer.js?commonjs-module"}]},{"name":"copy-text-to-clipboard/index.js?commonjs-module","uid":"ec5f2780-601"},{"name":"date-fns-tz","children":[{"name":"format","children":[{"uid":"ec5f2780-1311","name":"index.js?commonjs-module"},{"name":"formatters/index.js?commonjs-module","uid":"ec5f2780-1461"}]},{"name":"_lib","children":[{"name":"tzIntlTimeZoneName/index.js?commonjs-module","uid":"ec5f2780-1463"},{"name":"tzParseTimezone/index.js?commonjs-module","uid":"ec5f2780-1467"},{"name":"tzTokenizeDate/index.js?commonjs-module","uid":"ec5f2780-1469"},{"name":"newDateUTC/index.js?commonjs-module","uid":"ec5f2780-1473"},{"name":"tzPattern/index.js?commonjs-module","uid":"ec5f2780-1483"}]},{"name":"toDate/index.js?commonjs-module","uid":"ec5f2780-1481"},{"name":"formatInTimeZone/index.js?commonjs-module","uid":"ec5f2780-1491"},{"name":"utcToZonedTime/index.js?commonjs-module","uid":"ec5f2780-1501"},{"name":"getTimezoneOffset/index.js?commonjs-module","uid":"ec5f2780-1507"},{"name":"zonedTimeToUtc/index.js?commonjs-module","uid":"ec5f2780-1511"}]},{"name":"date-fns","children":[{"name":"format/index.js?commonjs-module","uid":"ec5f2780-1313"},{"name":"isValid/index.js?commonjs-module","uid":"ec5f2780-1319"},{"name":"isDate/index.js?commonjs-module","uid":"ec5f2780-1321"},{"name":"_lib","children":[{"name":"requiredArgs/index.js?commonjs-module","uid":"ec5f2780-1327"},{"name":"toInteger/index.js?commonjs-module","uid":"ec5f2780-1343"},{"name":"format","children":[{"name":"formatters/index.js?commonjs-module","uid":"ec5f2780-1351"},{"name":"lightFormatters/index.js?commonjs-module","uid":"ec5f2780-1397"},{"name":"longFormatters/index.js?commonjs-module","uid":"ec5f2780-1403"}]},{"name":"getUTCDayOfYear/index.js?commonjs-module","uid":"ec5f2780-1353"},{"name":"getUTCISOWeek/index.js?commonjs-module","uid":"ec5f2780-1357"},{"name":"startOfUTCISOWeek/index.js?commonjs-module","uid":"ec5f2780-1359"},{"name":"startOfUTCISOWeekYear/index.js?commonjs-module","uid":"ec5f2780-1363"},{"name":"getUTCISOWeekYear/index.js?commonjs-module","uid":"ec5f2780-1365"},{"name":"getUTCWeek/index.js?commonjs-module","uid":"ec5f2780-1373"},{"name":"startOfUTCWeek/index.js?commonjs-module","uid":"ec5f2780-1375"},{"name":"defaultOptions/index.js?commonjs-exports","uid":"ec5f2780-1377"},{"name":"startOfUTCWeekYear/index.js?commonjs-module","uid":"ec5f2780-1383"},{"name":"getUTCWeekYear/index.js?commonjs-module","uid":"ec5f2780-1385"},{"name":"addLeadingZeros/index.js?commonjs-module","uid":"ec5f2780-1393"},{"name":"getTimezoneOffsetInMilliseconds/index.js?commonjs-module","uid":"ec5f2780-1407"},{"name":"protectedTokens/index.js?commonjs-exports","uid":"ec5f2780-1411"},{"name":"defaultLocale/index.js?commonjs-module","uid":"eLine truncated
const run = () => {
const width = window.innerWidth;