diff --git a/MCP_README.md b/MCP_README.md
new file mode 100644
index 00000000..0d32fac1
--- /dev/null
+++ b/MCP_README.md
@@ -0,0 +1,152 @@
+# LTX-Desktop MCP Server
+
+An MCP (Model Context Protocol) server that wraps [LTX-Desktop](https://github.com/Lightricks/LTX-Desktop)'s AI video generation backend and project editing capabilities, enabling AI assistants to automate video creation and timeline editing.
+
+## Features
+
+| Capability | Description |
+|---|---|
+| 🎬 Video Generation | Text-to-video, image-to-video, audio-to-video |
+| 🖼️ Image Generation | Text-to-image with configurable resolution |
+| 🔄 Video Retake | AI re-generation of specific video segments |
+| 🎞️ Timeline Editing | Full timeline manipulation via project JSON |
+| ✂️ Clip Control | Trim, speed, reverse, opacity, volume |
+| 🎨 Color Grading | Brightness, contrast, saturation, temperature, exposure |
+| ✨ Effects | Blur, sharpen, glow, vignette, grain, LUT presets |
+| 🔤 Text & Subtitles | Text overlays and subtitle tracks |
+| 🔀 Transitions | Dissolve, fade, wipe (8 types) |
+| 📦 Export | FFmpeg-based video export (H.264, ProRes, VP9) |
+
+## Architecture
+
+```
+┌─────────────────┐     stdio      ┌──────────────────┐
+│   AI Assistant   │◄─────────────►│  MCP Server      │
+│ (Claude, etc.)   │               │  ltx_mcp_server.py│
+└─────────────────┘               └────┬────────┬─────┘
+                                       │        │
+                              HTTP :8000│        │HTTP :8100
+                                       ▼        ▼
+                                 ┌──────────┐ ┌──────────────┐
+                                 │ Python   │ │ Electron     │
+                                 │ Backend  │ │ Project      │
+                                 │ (FastAPI)│ │ Bridge       │
+                                 └──────────┘ └──────────────┘
+                                  Video Gen    Timeline JSON
+                                  Model Mgmt   Export/Import
+```
+
+## Prerequisites
+
+- [LTX-Desktop](https://github.com/Lightricks/LTX-Desktop) installed and running
+- Python 3.10+
+- NVIDIA GPU with ≥32GB VRAM (for local generation)
+
+## Installation
+
+```bash
+# 1. Install Python dependencies
+pip install mcp httpx
+
+# 2. Install LTX-Desktop dependencies
+cd /path/to/LTX-Desktop
+pnpm install
+
+# 3. Start LTX-Desktop
+pnpm dev
+```
+
+## Usage
+
+### Configure in your AI assistant
+
+Add to your MCP configuration (e.g. Claude Desktop `claude_desktop_config.json`):
+
+```json
+{
+  "mcpServers": {
+    "ltx-desktop": {
+      "command": "python3",
+      "args": ["/path/to/LTX-Desktop/ltx_mcp_server.py"],
+      "env": {
+        "LTX_BACKEND_URL": "http://127.0.0.1:8000",
+        "LTX_BRIDGE_URL": "http://127.0.0.1:8100"
+      }
+    }
+  }
+}
+```
+
+### Available MCP Tools
+
+#### Video Generation (Backend API)
+
+| Tool | Description |
+|---|---|
+| `generate_video` | Generate video from text/image/audio |
+| `generate_image` | Generate images from text |
+| `retake_video` | Re-generate a portion of a video |
+| `suggest_gap_prompt` | AI-suggest prompt for timeline gaps |
+| `get_generation_progress` | Check generation progress |
+| `cancel_generation` | Cancel running generation |
+
+#### Model Management
+
+| Tool | Description |
+|---|---|
+| `get_health` | Backend health & GPU status |
+| `get_models_status` | Model download status |
+| `download_models` | Trigger model downloads |
+
+#### Project & Timeline Editing (Electron Bridge)
+
+| Tool | Description |
+|---|---|
+| `list_projects` | List all projects |
+| `get_project` | Get full project JSON |
+| `update_project` | Update project (timeline, clips, etc.) |
+| `export_video` | Export timeline as video file |
+| `import_asset` | Import local file into project |
+
+### Example Workflow
+
+```
+User: "Generate 3 video clips about a sunset and arrange them on the timeline with dissolve transitions"
+
+AI Assistant:
+1. generate_video(prompt="golden sunset over ocean, waves crashing") → clip1.mp4
+2. generate_video(prompt="sun dipping below horizon, orange sky") → clip2.mp4  
+3. generate_video(prompt="twilight sky with first stars appearing") → clip3.mp4
+4. get_project(project_id) → read current project
+5. Add 3 assets + 3 clips to timeline with:
+   - clip1: startTime=0, transitionOut={type: "dissolve", duration: 0.5}
+   - clip2: startTime=5, transitionIn/Out={type: "dissolve", duration: 0.5}
+   - clip3: startTime=10, transitionIn={type: "dissolve", duration: 0.5}
+6. update_project(project_id, modified_json)
+7. export_video(project_id, "/home/user/sunset_montage.mp4")
+```
+
+## Environment Variables
+
+| Variable | Default | Description |
+|---|---|---|
+| `LTX_BACKEND_URL` | `http://127.0.0.1:8000` | LTX-Desktop FastAPI backend URL |
+| `LTX_BRIDGE_URL` | `http://127.0.0.1:8100` | Electron project bridge URL |
+| `LTX_TIMEOUT` | `600` | Request timeout in seconds |
+
+## Project Structure (MCP additions)
+
+```
+LTX-Desktop/
+├── ltx_mcp_server.py          # MCP Server (standalone script)
+├── electron/
+│   ├── project-bridge.ts      # HTTP bridge for project JSON access
+│   └── main.ts                # Modified: registers project bridge
+└── .agents/skills/
+    └── ltx-desktop-mcp/
+        └── SKILL.md            # AI assistant skill documentation
+```
+
+## License
+
+This MCP integration follows the same license as [LTX-Desktop](https://github.com/Lightricks/LTX-Desktop).
diff --git a/electron/main.ts b/electron/main.ts
index e79e0130..1b764c27 100644
--- a/electron/main.ts
+++ b/electron/main.ts
@@ -8,6 +8,7 @@ import { registerFileHandlers } from './ipc/file-handlers'
 import { registerLogHandlers } from './ipc/log-handlers'
 import { registerVideoProcessingHandlers } from './ipc/video-processing-handlers'
 import { initSessionLog } from './logging-management'
+import { startProjectBridge, stopProjectBridge } from './project-bridge'
 import { stopPythonBackend } from './python-backend'
 import { initAutoUpdater } from './updater'
 import { createWindow, getMainWindow } from './window'
@@ -56,6 +57,7 @@ if (!gotLock) {
     setupCSP()
     createWindow()
     initAutoUpdater()
+    startProjectBridge()
     // Python setup + backend start are now driven by the renderer via IPC
 
     // Fire analytics event (no-op if user hasn't opted in)
@@ -77,6 +79,7 @@ if (!gotLock) {
 
   app.on('before-quit', () => {
     stopExportProcess()
+    stopProjectBridge()
     stopPythonBackend()
   })
 }
diff --git a/electron/project-bridge.ts b/electron/project-bridge.ts
new file mode 100644
index 00000000..6e96103c
--- /dev/null
+++ b/electron/project-bridge.ts
@@ -0,0 +1,264 @@
+/**
+ * Project Bridge — lightweight HTTP server (port 8100) that exposes
+ * project data from the renderer's localStorage to external tools
+ * (MCP Server, scripts, etc.).
+ *
+ * Endpoints:
+ *   GET  /api/projects                — list all projects
+ *   GET  /api/projects/:id            — get a single project
+ *   PUT  /api/projects/:id            — update a single project
+ *   POST /api/export                  — trigger FFmpeg export
+ *   POST /api/import-asset            — copy a file into project assets
+ */
+
+import http from 'http'
+import { getMainWindow } from './window'
+import { logger } from './logger'
+
+const BRIDGE_PORT = 8100
+
+let bridgeServer: http.Server | null = null
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function jsonResponse(res: http.ServerResponse, status: number, body: unknown): void {
+  const payload = JSON.stringify(body)
+  res.writeHead(status, {
+    'Content-Type': 'application/json',
+    'Access-Control-Allow-Origin': '*',
+    'Access-Control-Allow-Methods': 'GET, PUT, POST, OPTIONS',
+    'Access-Control-Allow-Headers': 'Content-Type',
+  })
+  res.end(payload)
+}
+
+function readBody(req: http.IncomingMessage): Promise<string> {
+  return new Promise((resolve, reject) => {
+    const chunks: Buffer[] = []
+    req.on('data', (chunk: Buffer) => chunks.push(chunk))
+    req.on('end', () => resolve(Buffer.concat(chunks).toString('utf8')))
+    req.on('error', reject)
+  })
+}
+
+/** Execute JS in the renderer and return the result. */
+async function rendererEval<T>(js: string): Promise<T> {
+  const win = getMainWindow()
+  if (!win) throw new Error('Main window not available')
+  return win.webContents.executeJavaScript(js) as Promise<T>
+}
+
+// ---------------------------------------------------------------------------
+// Route handlers
+// ---------------------------------------------------------------------------
+
+async function handleListProjects(res: http.ServerResponse): Promise<void> {
+  try {
+    const projects = await rendererEval<unknown>(
+      `JSON.parse(localStorage.getItem('ltx-projects') || '[]')`
+    )
+    jsonResponse(res, 200, projects)
+  } catch (err) {
+    jsonResponse(res, 500, { error: String(err) })
+  }
+}
+
+async function handleGetProject(res: http.ServerResponse, projectId: string): Promise<void> {
+  try {
+    const projects = await rendererEval<Array<{ id: string }>>(
+      `JSON.parse(localStorage.getItem('ltx-projects') || '[]')`
+    )
+    const project = projects.find((p) => p.id === projectId)
+    if (!project) {
+      jsonResponse(res, 404, { error: `Project not found: ${projectId}` })
+      return
+    }
+    jsonResponse(res, 200, project)
+  } catch (err) {
+    jsonResponse(res, 500, { error: String(err) })
+  }
+}
+
+async function handleUpdateProject(
+  req: http.IncomingMessage,
+  res: http.ServerResponse,
+  projectId: string
+): Promise<void> {
+  try {
+    const body = await readBody(req)
+    const updatedProject = JSON.parse(body)
+
+    // Read current projects, replace the matching one
+    const projects = await rendererEval<Array<{ id: string }>>(
+      `JSON.parse(localStorage.getItem('ltx-projects') || '[]')`
+    )
+    const idx = projects.findIndex((p) => p.id === projectId)
+    if (idx === -1) {
+      jsonResponse(res, 404, { error: `Project not found: ${projectId}` })
+      return
+    }
+
+    // Ensure the project ID is preserved
+    updatedProject.id = projectId
+    updatedProject.updatedAt = Date.now()
+    projects[idx] = updatedProject
+
+    // Write back to localStorage
+    const escaped = JSON.stringify(JSON.stringify(projects))
+    await rendererEval<void>(`localStorage.setItem('ltx-projects', ${escaped})`)
+
+    // Trigger a storage event so React picks up the change
+    await rendererEval<void>(
+      `window.dispatchEvent(new StorageEvent('storage', { key: 'ltx-projects' }))`
+    )
+
+    // Force React re-render by dispatching a custom reload event
+    await rendererEval<void>(
+      `window.dispatchEvent(new CustomEvent('ltx-projects-reload'))`
+    )
+
+    jsonResponse(res, 200, { status: 'updated', id: projectId })
+  } catch (err) {
+    jsonResponse(res, 500, { error: String(err) })
+  }
+}
+
+async function handleExport(
+  req: http.IncomingMessage,
+  res: http.ServerResponse
+): Promise<void> {
+  try {
+    const body = await readBody(req)
+    const params = JSON.parse(body)
+
+    const { projectId, outputPath, width, height, fps, codec, quality } = params
+
+    // Read the project and active timeline
+    const projects = await rendererEval<Array<{ id: string; timelines?: Array<{ id: string; clips?: unknown[] }>; activeTimelineId?: string }>>(
+      `JSON.parse(localStorage.getItem('ltx-projects') || '[]')`
+    )
+    const project = projects.find((p) => p.id === projectId)
+    if (!project) {
+      jsonResponse(res, 404, { error: `Project not found: ${projectId}` })
+      return
+    }
+
+    const timeline = project.timelines?.find((t) => t.id === project.activeTimelineId) || project.timelines?.[0]
+    if (!timeline || !timeline.clips?.length) {
+      jsonResponse(res, 400, { error: 'No clips in timeline to export' })
+      return
+    }
+
+    // Delegate to the existing export-native IPC handler by invoking it via renderer
+    const exportData = {
+      clips: timeline.clips,
+      outputPath,
+      codec: codec || 'h264',
+      width: width || 1920,
+      height: height || 1080,
+      fps: fps || 24,
+      quality: quality || 80,
+    }
+
+    const escaped = JSON.stringify(JSON.stringify(exportData))
+    const result = await rendererEval<{ success?: boolean; error?: string }>(
+      `window.electronAPI.exportNative(JSON.parse(${escaped}))`
+    )
+    jsonResponse(res, result?.success ? 200 : 500, result)
+  } catch (err) {
+    jsonResponse(res, 500, { error: String(err) })
+  }
+}
+
+async function handleImportAsset(
+  req: http.IncomingMessage,
+  res: http.ServerResponse
+): Promise<void> {
+  try {
+    const body = await readBody(req)
+    const { projectId, filePath } = JSON.parse(body)
+
+    if (!projectId || !filePath) {
+      jsonResponse(res, 400, { error: 'projectId and filePath are required' })
+      return
+    }
+
+    // Use the existing copyToProjectAssets IPC handler via renderer
+    const escaped1 = JSON.stringify(filePath)
+    const escaped2 = JSON.stringify(projectId)
+    const result = await rendererEval<{ success: boolean; path?: string; url?: string; error?: string }>(
+      `window.electronAPI.copyToProjectAssets(${escaped1}, ${escaped2})`
+    )
+    jsonResponse(res, result?.success ? 200 : 500, result)
+  } catch (err) {
+    jsonResponse(res, 500, { error: String(err) })
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Request router
+// ---------------------------------------------------------------------------
+
+async function handleRequest(
+  req: http.IncomingMessage,
+  res: http.ServerResponse
+): Promise<void> {
+  const method = req.method?.toUpperCase() || 'GET'
+  const url = req.url || '/'
+
+  // CORS preflight
+  if (method === 'OPTIONS') {
+    jsonResponse(res, 204, '')
+    return
+  }
+
+  // Route matching
+  const projectMatch = url.match(/^\/api\/projects\/([^/]+)$/)
+
+  if (url === '/api/projects' && method === 'GET') {
+    await handleListProjects(res)
+  } else if (projectMatch && method === 'GET') {
+    await handleGetProject(res, projectMatch[1])
+  } else if (projectMatch && method === 'PUT') {
+    await handleUpdateProject(req, res, projectMatch[1])
+  } else if (url === '/api/export' && method === 'POST') {
+    await handleExport(req, res)
+  } else if (url === '/api/import-asset' && method === 'POST') {
+    await handleImportAsset(req, res)
+  } else {
+    jsonResponse(res, 404, { error: `Not found: ${method} ${url}` })
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Public API
+// ---------------------------------------------------------------------------
+
+export function startProjectBridge(): void {
+  if (bridgeServer) return
+
+  bridgeServer = http.createServer((req, res) => {
+    handleRequest(req, res).catch((err) => {
+      logger.error(`[ProjectBridge] Unhandled error: ${err}`)
+      jsonResponse(res, 500, { error: 'Internal server error' })
+    })
+  })
+
+  bridgeServer.listen(BRIDGE_PORT, '127.0.0.1', () => {
+    logger.info(`[ProjectBridge] Listening on http://127.0.0.1:${BRIDGE_PORT}`)
+  })
+
+  bridgeServer.on('error', (err) => {
+    logger.error(`[ProjectBridge] Server error: ${err}`)
+  })
+}
+
+export function stopProjectBridge(): void {
+  if (bridgeServer) {
+    bridgeServer.close()
+    bridgeServer = null
+    logger.info('[ProjectBridge] Stopped')
+  }
+}
diff --git a/ltx-desktop-mcp/SKILL.md b/ltx-desktop-mcp/SKILL.md
new file mode 100644
index 00000000..f891a368
--- /dev/null
+++ b/ltx-desktop-mcp/SKILL.md
@@ -0,0 +1,686 @@
+---
+name: LTX-Desktop MCP 视频生成与编辑
+description: 使用 LTX-Desktop MCP Server 进行 AI 视频生成、时间轴编辑和视频导出的完整指南
+---
+
+# LTX-Desktop MCP 视频生成与编辑
+
+## 概述
+
+LTX-Desktop 是一个基于 AI 的视频生成和编辑桌面应用。通过 MCP Server，AI 助手可以：
+- **生成视频/图片**：调用后端 API（端口 8000）
+- **编辑时间轴**：通过 Electron 桥（端口 8100）读写项目 JSON
+- **导出成片**：调用 FFmpeg 合成导出
+
+## 前置条件
+
+1. LTX-Desktop 必须已启动（`pnpm dev`），确保后端（:8000）和桥（:8100）都在运行
+2. MCP Server 脚本路径：`/home/ve/下载/LTX-Desktop/ltx_mcp_server.py`
+
+---
+
+## MCP 工具详细参数说明
+
+### 一、视频生成工具
+
+#### `generate_video` — 生成视频
+
+从文本提示词生成视频，或基于输入图片/音频生成视频。
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `prompt` | string | ✅ 必填 | — | 视频内容描述。不能为空，前后空格会被自动去除 |
+| `resolution` | string | 选填 | `"512p"` | 生成分辨率 |
+| `model` | string | 选填 | `"fast"` | 模型变体。`"fast"` = 快速蒸馏模型，`"pro"` = 高质量模型 |
+| `duration` | string | 选填 | `"2"` | 视频时长（秒），传字符串 |
+| `fps` | string | 选填 | `"24"` | 帧率，传字符串 |
+| `camera_motion` | string | 选填 | `"none"` | 摄像机运动方式，见下表 |
+| `negative_prompt` | string | 选填 | `""` | 负面提示词，描述要避免的内容 |
+| `audio` | string | 选填 | `"false"` | 是否同时生成音频。`"true"` 或 `"false"` |
+| `image_path` | string | 选填 | `null` | 输入图片的**本地绝对路径**，用于图生视频（image-to-video） |
+| `audio_path` | string | 选填 | `null` | 输入音频的**本地绝对路径**，用于音生视频（audio-to-video） |
+| `aspect_ratio` | string | 选填 | `"16:9"` | 画面比例。仅支持 `"16:9"` 或 `"9:16"` |
+
+**`camera_motion` 可选值：**
+
+| 值 | 说明 |
+|---|---|
+| `"none"` | 无运动（默认） |
+| `"static"` | 完全静止镜头 |
+| `"dolly_in"` | 推镜头（向前移动） |
+| `"dolly_out"` | 拉镜头（向后移动） |
+| `"dolly_left"` | 左移 |
+| `"dolly_right"` | 右移 |
+| `"jib_up"` | 镜头上升 |
+| `"jib_down"` | 镜头下降 |
+| `"focus_shift"` | 焦点转移 |
+
+**返回值：**
+```json
+{ "status": "complete", "video_path": "/absolute/path/to/output.mp4" }
+```
+
+---
+
+#### `generate_image` — 生成图片
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `prompt` | string | ✅ 必填 | — | 图片内容描述。不能为空 |
+| `width` | int | 选填 | `1024` | 图片宽度（像素） |
+| `height` | int | 选填 | `1024` | 图片高度（像素） |
+| `num_steps` | int | 选填 | `4` | 扩散步数，越大质量越高但越慢 |
+| `num_images` | int | 选填 | `1` | 生成图片数量 |
+
+**返回值：**
+```json
+{ "status": "complete", "image_paths": ["/path/to/img1.png", "/path/to/img2.png"] }
+```
+
+---
+
+#### `retake_video` — 视频局部重拍
+
+对已有视频的某个时间段进行 AI 重新生成。
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `video_path` | string | ✅ 必填 | — | 源视频的本地绝对路径 |
+| `start_time` | float | ✅ 必填 | — | 重拍起始时间（秒），如 `1.5` |
+| `duration` | float | ✅ 必填 | — | 重拍时长（秒），如 `2.0` |
+| `prompt` | string | 选填 | `""` | 重拍内容的提示词，为空则保持相似风格 |
+| `mode` | string | 选填 | `"replace_audio_and_video"` | 重拍模式，同时替换音视频 |
+
+**返回值：**
+```json
+{ "status": "complete", "video_path": "/path/to/retake_output.mp4" }
+```
+
+---
+
+#### `suggest_gap_prompt` — AI 建议补帧提示词
+
+当时间轴上两段片段之间有空隙时，AI 根据上下文建议一个合适的提示词来生成过渡内容。
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `before_prompt` | string | 选填 | `""` | 空隙前面那段片段的提示词 |
+| `after_prompt` | string | 选填 | `""` | 空隙后面那段片段的提示词 |
+| `gap_duration` | float | 选填 | `5` | 空隙时长（秒） |
+| `mode` | string | 选填 | `"t2v"` | 生成模式。`"t2v"` = 文生视频 |
+| `before_frame` | string | 选填 | `null` | 空隙前最后一帧图片的路径 |
+| `after_frame` | string | 选填 | `null` | 空隙后第一帧图片的路径 |
+| `input_image` | string | 选填 | `null` | 输入参考图片路径 |
+
+**返回值：**
+```json
+{ "status": "success", "suggested_prompt": "A smooth camera pan transitioning from..." }
+```
+
+---
+
+### 二、生成控制工具
+
+#### `get_generation_progress` — 查询生成进度
+
+无参数。实时查询当前视频/图片生成的进度。
+
+**返回值：**
+```json
+{
+  "status": "running",       // "idle" | "running" | "complete" | "cancelled" | "error"
+  "phase": "denoising",      // 当前阶段描述
+  "progress": 45,            // 总进度百分比 0-100
+  "currentStep": 9,          // 当前扩散步数
+  "totalSteps": 20           // 总扩散步数
+}
+```
+
+---
+
+#### `cancel_generation` — 取消生成
+
+无参数。取消当前正在运行的生成任务。
+
+**返回值：**
+```json
+{ "status": "cancelled", "id": "gen-xxx" }
+```
+
+---
+
+### 三、模型管理工具
+
+#### `get_health` — 健康检查
+
+无参数。检查后端状态和 GPU 信息。
+
+**返回值：**
+```json
+{
+  "status": "ready",
+  "models_loaded": true,
+  "active_model": "fast",
+  "gpu_info": { "name": "NVIDIA GB202", "vram": 131072, "vramUsed": 45000 },
+  "sage_attention": true,
+  "models_status": [
+    { "id": "checkpoint", "name": "LTX Checkpoint", "loaded": true, "downloaded": true }
+  ]
+}
+```
+
+---
+
+#### `get_models_status` — 模型下载状态
+
+无参数。查看所有模型文件的下载状态和大小。
+
+**返回值：**
+```json
+{
+  "all_downloaded": false,
+  "total_size_gb": 98.5,
+  "downloaded_size_gb": 45.2,
+  "models_path": "/home/ve/.local/share/LTXDesktop/models",
+  "has_api_key": true,
+  "models": [
+    {
+      "id": "checkpoint",
+      "name": "LTX Video Checkpoint",
+      "downloaded": true,
+      "size": 12345678900,
+      "expected_size": 12345678900,
+      "required": true
+    }
+  ]
+}
+```
+
+---
+
+#### `download_models` — 触发模型下载
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `model_types` | list[string] | 选填 | `null`（全部下载） | 要下载的模型类型列表 |
+
+**`model_types` 可选值：**
+
+| 值 | 说明 |
+|---|---|
+| `"checkpoint"` | 主模型检查点（最大，~15GB） |
+| `"upsampler"` | 超分辨率上采样器 |
+| `"distilled_lora"` | 蒸馏 LoRA（用于 fast 模式加速） |
+| `"ic_lora"` | 图像条件 LoRA |
+| `"depth_processor"` | 深度估计处理器 |
+| `"person_detector"` | 人体检测器 |
+| `"pose_processor"` | 姿态估计处理器 |
+| `"text_encoder"` | 文本编码器（~10GB，本地文本编码用） |
+| `"zit"` | ZIT 图像修复模型 |
+
+---
+
+### 四、项目管理工具（Electron 桥）
+
+#### `list_projects` — 列出所有项目
+
+无参数。返回所有项目的摘要信息。
+
+**返回值：** `Project[]` 数组（见下方 JSON 结构定义）
+
+---
+
+#### `get_project` — 获取项目完整数据
+
+| 参数 | 类型 | 必填 | 说明 |
+|---|---|---|---|
+| `project_id` | string | ✅ 必填 | 项目 ID，格式如 `"project-1710000000000-abc123def"` |
+
+**返回值：** 完整的 `Project` 对象 JSON（包含 assets、timelines、clips 等全部数据）
+
+---
+
+#### `update_project` — 更新项目
+
+| 参数 | 类型 | 必填 | 说明 |
+|---|---|---|---|
+| `project_id` | string | ✅ 必填 | 项目 ID |
+| `project_json` | string | ✅ 必填 | **完整的项目 JSON 字符串**。先用 `get_project` 读取，修改后传回 |
+
+> ⚠️ **重要**：必须传完整的 Project JSON，不支持局部更新。先读后改再写！
+
+---
+
+#### `export_video` — 导出视频
+
+| 参数 | 类型 | 必填 | 默认值 | 说明 |
+|---|---|---|---|---|
+| `project_id` | string | ✅ 必填 | — | 要导出的项目 ID |
+| `output_path` | string | ✅ 必填 | — | 输出文件的本地绝对路径（如 `/home/ve/output.mp4`） |
+| `width` | int | 选填 | `1920` | 输出视频宽度（像素） |
+| `height` | int | 选填 | `1080` | 输出视频高度（像素） |
+| `fps` | int | 选填 | `24` | 输出帧率 |
+| `codec` | string | 选填 | `"h264"` | 编码器。`"h264"` / `"prores"` / `"vp9"` |
+| `quality` | int | 选填 | `80` | 输出质量。h264: CRF 值(0-51, 越小越好)；prores: profile(0-3)；vp9: 比特率(MB) |
+
+---
+
+#### `import_asset` — 导入素材
+
+| 参数 | 类型 | 必填 | 说明 |
+|---|---|---|---|
+| `project_id` | string | ✅ 必填 | 目标项目 ID |
+| `file_path` | string | ✅ 必填 | 要导入的本地文件绝对路径（视频/图片/音频） |
+
+**返回值：**
+```json
+{ "success": true, "path": "/copied/path/to/file.mp4", "url": "file:///copied/path/to/file.mp4" }
+```
+
+---
+
+## 项目 JSON 完整结构定义
+
+### Project（项目）
+
+```json
+{
+  "id": "project-{timestamp}-{random9}",
+  "name": "项目名称",
+  "createdAt": 1710000000000,
+  "updatedAt": 1710000000000,
+  "thumbnail": "file:///path/to/thumb.png",
+  "assets": [ /* Asset[] */ ],
+  "timelines": [ /* Timeline[] */ ],
+  "activeTimelineId": "timeline-xxx"
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `id` | string | 唯一 ID，格式 `project-{timestamp}-{random9}` |
+| `name` | string | 项目名称 |
+| `createdAt` | number | 创建时间（毫秒时间戳） |
+| `updatedAt` | number | 最后更新时间（毫秒时间戳） |
+| `thumbnail` | string? | 项目缩略图 URL |
+| `assets` | Asset[] | 项目中的所有素材 |
+| `timelines` | Timeline[] | 时间轴列表（可多条） |
+| `activeTimelineId` | string? | 当前激活的时间轴 ID |
+
+---
+
+### Asset（素材）
+
+```json
+{
+  "id": "asset-{timestamp}-{random9}",
+  "type": "video",
+  "path": "/absolute/path/to/file.mp4",
+  "url": "file:///absolute/path/to/file.mp4",
+  "prompt": "生成时的提示词",
+  "resolution": "512p",
+  "duration": 5.0,
+  "createdAt": 1710000000000,
+  "thumbnail": "file:///path/to/thumb.png",
+  "favorite": false,
+  "takes": [],
+  "activeTakeIndex": 0
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `id` | string | 唯一 ID |
+| `type` | string | `"video"` / `"image"` / `"audio"` / `"adjustment"` |
+| `path` | string | 文件在磁盘上的绝对路径 |
+| `url` | string | `file://` 协议 URL |
+| `prompt` | string? | AI 生成时的提示词 |
+| `resolution` | string? | 生成分辨率 |
+| `duration` | number? | 素材时长（秒），图片无此字段 |
+| `createdAt` | number | 创建时间戳 |
+| `thumbnail` | string? | 缩略图 URL |
+| `favorite` | boolean | 是否收藏 |
+| `takes` | AssetTake[] | 重拍/多版本历史 |
+| `activeTakeIndex` | number? | 当前活跃版本索引 |
+
+---
+
+### Timeline（时间轴）
+
+```json
+{
+  "id": "timeline-{timestamp}-{random9}",
+  "name": "Timeline 1",
+  "createdAt": 1710000000000,
+  "tracks": [
+    { "id": "track-v1", "name": "V1", "kind": "video", "muted": false, "locked": false },
+    { "id": "track-v2", "name": "V2", "kind": "video", "muted": false, "locked": false },
+    { "id": "track-v3", "name": "V3", "kind": "video", "muted": false, "locked": false },
+    { "id": "track-a1", "name": "A1", "kind": "audio", "muted": false, "locked": false },
+    { "id": "track-a2", "name": "A2", "kind": "audio", "muted": false, "locked": false }
+  ],
+  "clips": [],
+  "subtitles": []
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `tracks[].kind` | string | `"video"` 或 `"audio"` |
+| `tracks[].muted` | boolean | 轨道是否静音 |
+| `tracks[].locked` | boolean | 轨道是否锁定（锁定后不可编辑） |
+
+---
+
+### TimelineClip（片段）— 核心数据结构
+
+```json
+{
+  "id": "clip-{timestamp}-{random9}",
+  "assetId": "asset-xxx",
+  "type": "video",
+  "trackIndex": 0,
+  "startTime": 0,
+  "duration": 5.0,
+  "trimStart": 0,
+  "trimEnd": 0,
+  "speed": 1.0,
+  "reversed": false,
+  "muted": false,
+  "volume": 100,
+  "opacity": 100,
+  "flipH": false,
+  "flipV": false,
+  "linkedClipIds": [],
+  "transitionIn": { "type": "none", "duration": 0 },
+  "transitionOut": { "type": "none", "duration": 0 },
+  "colorCorrection": {
+    "brightness": 0, "contrast": 0, "saturation": 0,
+    "temperature": 0, "tint": 0, "exposure": 0,
+    "highlights": 0, "shadows": 0
+  },
+  "effects": [],
+  "asset": null,
+  "textStyle": null
+}
+```
+
+**时间与播放控制：**
+
+| 字段 | 类型 | 范围 | 说明 |
+|---|---|---|---|
+| `type` | string | — | `"video"` / `"image"` / `"audio"` / `"adjustment"` / `"text"` |
+| `trackIndex` | int | 0-4 | 轨道索引。**V1=0, V2=1, V3=2, A1=3, A2=4** |
+| `startTime` | float | ≥0 | 在时间轴上的起始时间（秒） |
+| `duration` | float | >0 | 播放时长（秒） |
+| `trimStart` | float | ≥0 | 从素材头部裁掉的秒数 |
+| `trimEnd` | float | ≥0 | 从素材尾部裁掉的秒数 |
+| `speed` | float | >0 | 播放速度。`0.25`=四分之一速, `0.5`=慢放, `1.0`=正常, `2.0`=快进, `4.0`=四倍速 |
+| `reversed` | boolean | — | `true` = 倒放 |
+
+**音频控制：**
+
+| 字段 | 类型 | 范围 | 说明 |
+|---|---|---|---|
+| `muted` | boolean | — | `true` = 静音 |
+| `volume` | int | 0-100 | 音量百分比 |
+
+**视觉控制：**
+
+| 字段 | 类型 | 范围 | 说明 |
+|---|---|---|---|
+| `opacity` | int | 0-100 | 透明度。100=不透明, 0=完全透明 |
+| `flipH` | boolean | — | 水平翻转 |
+| `flipV` | boolean | — | 垂直翻转 |
+
+**关联：**
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `assetId` | string | 引用的素材 ID |
+| `linkedClipIds` | string[] | 关联的片段 ID 列表（音视频联动，移动一个另一个跟着） |
+| `asset` | object? | 可设为 `null`，前端会根据 `assetId` 自动关联 |
+
+---
+
+### Transition（转场）
+
+`transitionIn` 和 `transitionOut` 字段的结构：
+
+```json
+{ "type": "dissolve", "duration": 0.5 }
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `type` | string | 转场类型，见下表 |
+| `duration` | float | 转场时长（秒），通常 0.3-1.5 |
+
+**转场类型：**
+
+| 值 | 效果 |
+|---|---|
+| `"none"` | 无转场（硬切） |
+| `"dissolve"` | 溶解/交叉淡化 |
+| `"fade-to-black"` | 淡入/淡出黑色 |
+| `"fade-to-white"` | 淡入/淡出白色 |
+| `"wipe-left"` | 从右向左擦除 |
+| `"wipe-right"` | 从左向右擦除 |
+| `"wipe-up"` | 从下向上擦除 |
+| `"wipe-down"` | 从上向下擦除 |
+
+---
+
+### ColorCorrection（调色）
+
+所有值范围 **-100 到 100**，默认 **0**。
+
+| 字段 | 说明 |
+|---|---|
+| `brightness` | 亮度。正值提亮，负值压暗 |
+| `contrast` | 对比度。正值增强对比，负值降低 |
+| `saturation` | 饱和度。正值增强色彩，负值减弱，-100=黑白 |
+| `temperature` | 色温。正值偏暖（黄/橙），负值偏冷（蓝） |
+| `tint` | 色调偏移。正值偏品红，负值偏绿 |
+| `exposure` | 曝光。正值过曝，负值欠曝 |
+| `highlights` | 高光。正值提高高光区域亮度，负值压低 |
+| `shadows` | 阴影。正值提亮暗部，负值压暗 |
+
+---
+
+### Effects（特效）
+
+`effects` 是一个数组，每个元素：
+
+```json
+{
+  "type": "blur",
+  "intensity": 50,
+  "enabled": true,
+  "mask": null
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `type` | string | 特效类型，见下表 |
+| `intensity` | int | 强度 0-100 |
+| `enabled` | boolean | 是否启用 |
+| `mask` | object? | 可选遮罩（限定特效作用区域） |
+
+**特效类型：**
+
+| 值 | 效果 |
+|---|---|
+| `"blur"` | 高斯模糊 |
+| `"sharpen"` | 锐化 |
+| `"glow"` | 辉光/发光 |
+| `"vignette"` | 暗角（四周变暗） |
+| `"grain"` | 胶片颗粒噪点 |
+| `"lut-cinematic"` | LUT 预设：电影感 |
+| `"lut-vintage"` | LUT 预设：复古 |
+| `"lut-bw"` | LUT 预设：黑白 |
+| `"lut-cool"` | LUT 预设：冷色调 |
+| `"lut-warm"` | LUT 预设：暖色调 |
+| `"lut-muted"` | LUT 预设：低饱和 |
+| `"lut-vivid"` | LUT 预设：高饱和鲜艳 |
+
+**遮罩（Mask）结构：**
+
+```json
+{
+  "shape": "rectangle",
+  "x": 50, "y": 50,
+  "width": 30, "height": 30,
+  "rotation": 0,
+  "feather": 10,
+  "inverted": false
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `shape` | string | `"rectangle"` 或 `"ellipse"` |
+| `x`, `y` | float | 中心点位置（百分比 0-100） |
+| `width`, `height` | float | 尺寸（百分比 0-100） |
+| `rotation` | float | 旋转角度（度） |
+| `feather` | float | 边缘羽化程度 0-100 |
+| `inverted` | boolean | `true` = 反转遮罩（内部不受影响，外部受影响） |
+
+---
+
+### SubtitleClip（字幕）
+
+```json
+{
+  "id": "sub-{timestamp}-{random9}",
+  "text": "字幕内容",
+  "startTime": 1.0,
+  "endTime": 4.0,
+  "trackIndex": 0,
+  "style": {
+    "fontSize": 32,
+    "fontFamily": "sans-serif",
+    "fontWeight": "normal",
+    "color": "#FFFFFF",
+    "backgroundColor": "transparent",
+    "position": "bottom",
+    "italic": false
+  }
+}
+```
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `text` | string | 字幕文字内容 |
+| `startTime` | float | 显示起始时间（秒） |
+| `endTime` | float | 显示结束时间（秒） |
+| `trackIndex` | int | 字幕轨道索引 |
+| `style.fontSize` | int | 字号（像素） |
+| `style.fontFamily` | string | 字体，如 `"sans-serif"`, `"Inter"`, `"Arial"` |
+| `style.fontWeight` | string | 字重。`"normal"` / `"bold"` |
+| `style.color` | string | 文字颜色（十六进制如 `"#FFFFFF"`） |
+| `style.backgroundColor` | string | 背景色。`"transparent"` = 无背景 |
+| `style.position` | string | 显示位置。`"top"` / `"center"` / `"bottom"` |
+| `style.italic` | boolean | 是否斜体 |
+
+---
+
+### TextStyle（文字叠加）
+
+用于 `type: "text"` 的片段，放在 `clip.textStyle` 字段中：
+
+```json
+{
+  "text": "标题文字",
+  "fontSize": 64,
+  "fontFamily": "Inter, Arial, sans-serif",
+  "fontWeight": "bold",
+  "color": "#FFFFFF",
+  "backgroundColor": "transparent",
+  "positionX": 50,
+  "positionY": 50,
+  "opacity": 100,
+  "strokeColor": "#000000",
+  "strokeWidth": 2,
+  "shadowColor": "rgba(0,0,0,0.5)",
+  "shadowBlur": 4,
+  "letterSpacing": 0,
+  "lineHeight": 1.2
+}
+```
+
+| 字段 | 类型 | 范围 | 说明 |
+|---|---|---|---|
+| `text` | string | — | 显示的文字内容 |
+| `fontSize` | int | >0 | 字号（像素） |
+| `fontFamily` | string | — | 字体族 |
+| `fontWeight` | string | — | `"normal"` / `"bold"` / `"100"`-`"900"` |
+| `color` | string | — | 文字颜色（十六进制） |
+| `backgroundColor` | string | — | 背景色 |
+| `positionX` | float | 0-100 | 水平位置（百分比，50=居中） |
+| `positionY` | float | 0-100 | 垂直位置（百分比，50=居中） |
+| `opacity` | int | 0-100 | 透明度 |
+| `strokeColor` | string | — | 描边颜色 |
+| `strokeWidth` | float | ≥0 | 描边宽度 |
+| `shadowColor` | string | — | 阴影颜色 |
+| `shadowBlur` | float | ≥0 | 阴影模糊半径 |
+| `letterSpacing` | float | — | 字间距（像素） |
+| `lineHeight` | float | >0 | 行高倍数 |
+
+---
+
+## 常用操作配方
+
+### 1. 生成视频并添加到时间轴
+
+```
+步骤:
+1. generate_video(prompt="一只猫在沙滩上奔跑") → {"video_path": "/path/to/video.mp4"}
+2. get_project(project_id) → 拿到当前 project JSON
+3. 构造 Asset 对象，插入 project.assets 数组头部
+4. 构造 TimelineClip 对象，设置 assetId 指向刚才的 Asset
+5. 计算 startTime（= 时间轴上已有片段的末尾时间）
+6. 将 clip 插入 project.timelines[activeTimeline].clips
+7. update_project(project_id, 修改后的 JSON)
+```
+
+### 2. 音视频对齐
+
+```
+让视频 clip 和音频 clip 的 startTime 设为相同值，
+双方的 linkedClipIds 中互相引用对方的 id。
+```
+
+### 3. 顺序排列多个片段
+
+```
+clip1.startTime = 0
+clip2.startTime = clip1.startTime + clip1.duration
+clip3.startTime = clip2.startTime + clip2.duration
+```
+
+### 4. 生成 ID 的规则
+
+```
+timestamp = Date.now() 毫秒时间戳
+random9 = Math.random().toString(36).substr(2, 9) 的随机字符串
+
+Project:  "project-{timestamp}-{random9}"
+Asset:    "asset-{timestamp}-{random9}"
+Timeline: "timeline-{timestamp}-{random9}"
+Clip:     "clip-{timestamp}-{random9}"
+Subtitle: "sub-{timestamp}-{random9}"
+```
+
+---
+
+## 注意事项
+
+1. **先读后改**：修改项目前务必先 `get_project` 读取最新状态
+2. **ID 不可变**：不要修改已有对象的 `id` 字段
+3. **asset 字段**：clip 中的 `asset` 字段设为 `null` 即可，前端根据 `assetId` 自动关联
+4. **文件路径**：所有路径必须是绝对路径，URL 使用 `file://` 协议
+5. **GPU 要求**：视频/图片生成需要 NVIDIA GPU（≥32GB VRAM）+ 已下载模型
+6. **FFmpeg 要求**：视频导出依赖系统 FFmpeg
+7. **单次生成**：后端同一时刻只能运行一个生成任务，需等上一个完成或取消后才能开始下一个
diff --git a/ltx_mcp_server.py b/ltx_mcp_server.py
new file mode 100644
index 00000000..2419484e
--- /dev/null
+++ b/ltx_mcp_server.py
@@ -0,0 +1,383 @@
+#!/usr/bin/env python3
+"""LTX-Desktop MCP Server.
+
+Exposes LTX-Desktop's video generation backend (FastAPI on :8000) and
+Electron project bridge (HTTP on :8100) as MCP tools so AI assistants
+can automate video generation and timeline editing.
+
+Usage:
+    1. Start LTX-Desktop normally (pnpm dev)
+    2. Run this server:  python ltx_mcp_server.py
+    3. Register in your AI assistant's MCP configuration.
+"""
+from __future__ import annotations
+
+import os
+os.environ["PYTHONUNBUFFERED"] = "1"
+
+import json
+import logging
+from typing import Any
+
+import httpx
+from mcp.server.fastmcp import FastMCP
+
+# ---------------------------------------------------------------------------
+# Config
+# ---------------------------------------------------------------------------
+LTX_BACKEND = os.getenv("LTX_BACKEND_URL", "http://127.0.0.1:8000")
+LTX_BRIDGE  = os.getenv("LTX_BRIDGE_URL",  "http://127.0.0.1:8100")
+REQUEST_TIMEOUT = float(os.getenv("LTX_TIMEOUT", "600"))  # seconds
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger("ltx-mcp")
+
+mcp = FastMCP(
+    "ltx-desktop",
+    description="MCP server for LTX-Desktop: AI video generation + timeline editing",
+)
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+async def _backend_get(path: str, params: dict | None = None) -> dict[str, Any]:
+    """GET request to the LTX FastAPI backend."""
+    async with httpx.AsyncClient(timeout=REQUEST_TIMEOUT) as client:
+        resp = await client.get(f"{LTX_BACKEND}{path}", params=params)
+        resp.raise_for_status()
+        return resp.json()
+
+
+async def _backend_post(path: str, body: dict | None = None) -> dict[str, Any]:
+    """POST request to the LTX FastAPI backend."""
+    async with httpx.AsyncClient(timeout=REQUEST_TIMEOUT) as client:
+        resp = await client.post(f"{LTX_BACKEND}{path}", json=body or {})
+        resp.raise_for_status()
+        return resp.json()
+
+
+async def _bridge_get(path: str) -> Any:
+    """GET request to the Electron project bridge."""
+    async with httpx.AsyncClient(timeout=30) as client:
+        resp = await client.get(f"{LTX_BRIDGE}{path}")
+        resp.raise_for_status()
+        return resp.json()
+
+
+async def _bridge_post(path: str, body: dict | None = None) -> Any:
+    """POST / PUT request to the Electron project bridge."""
+    async with httpx.AsyncClient(timeout=60) as client:
+        resp = await client.post(f"{LTX_BRIDGE}{path}", json=body or {})
+        resp.raise_for_status()
+        return resp.json()
+
+
+async def _bridge_put(path: str, body: Any = None) -> Any:
+    """PUT request to the Electron project bridge."""
+    async with httpx.AsyncClient(timeout=60) as client:
+        resp = await client.put(f"{LTX_BRIDGE}{path}", json=body)
+        resp.raise_for_status()
+        return resp.json()
+
+
+# ===================================================================
+# BACKEND API TOOLS  (connect to FastAPI on :8000)
+# ===================================================================
+
+@mcp.tool()
+async def get_health() -> str:
+    """Check LTX-Desktop backend health: GPU status, loaded models, etc."""
+    result = await _backend_get("/health")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def get_models_status() -> str:
+    """Get the download status of all required model files (checkpoint, upsampler, text encoder, etc.)."""
+    result = await _backend_get("/api/models/status")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def download_models(model_types: list[str] | None = None) -> str:
+    """Start downloading required model files.
+
+    Args:
+        model_types: Optional list of model types to download.
+                     Valid values: checkpoint, upsampler, distilled_lora, ic_lora,
+                     depth_processor, person_detector, pose_processor, text_encoder, zit.
+                     If empty, downloads all required models.
+    """
+    body: dict[str, Any] = {}
+    if model_types:
+        body["modelTypes"] = model_types
+    result = await _backend_post("/api/models/download", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def generate_video(
+    prompt: str,
+    resolution: str = "512p",
+    model: str = "fast",
+    duration: str = "2",
+    fps: str = "24",
+    camera_motion: str = "none",
+    negative_prompt: str = "",
+    audio: str = "false",
+    image_path: str | None = None,
+    audio_path: str | None = None,
+    aspect_ratio: str = "16:9",
+) -> str:
+    """Generate a video from a text prompt (text-to-video), an image (image-to-video), or audio (audio-to-video).
+
+    Args:
+        prompt: The text description of the video to generate. Required.
+        resolution: Video resolution. Default "512p".
+        model: Model variant. "fast" (default) or "pro".
+        duration: Video duration in seconds. Default "2".
+        fps: Frames per second. Default "24".
+        camera_motion: Camera movement type. Options: none, dolly_in, dolly_out,
+                       dolly_left, dolly_right, jib_up, jib_down, static, focus_shift.
+        negative_prompt: Things to avoid in the generated video.
+        audio: Whether to generate audio. "true" or "false" (default).
+        image_path: Local file path of an input image for image-to-video generation.
+        audio_path: Local file path of an input audio for audio-to-video generation.
+        aspect_ratio: "16:9" (default) or "9:16".
+    """
+    body: dict[str, Any] = {
+        "prompt": prompt,
+        "resolution": resolution,
+        "model": model,
+        "duration": duration,
+        "fps": fps,
+        "cameraMotion": camera_motion,
+        "negativePrompt": negative_prompt,
+        "audio": audio,
+        "aspectRatio": aspect_ratio,
+    }
+    if image_path:
+        body["imagePath"] = image_path
+    if audio_path:
+        body["audioPath"] = audio_path
+
+    result = await _backend_post("/api/generate", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def generate_image(
+    prompt: str,
+    width: int = 1024,
+    height: int = 1024,
+    num_steps: int = 4,
+    num_images: int = 1,
+) -> str:
+    """Generate images from a text prompt.
+
+    Args:
+        prompt: The text description of the image to generate. Required.
+        width: Image width in pixels. Default 1024.
+        height: Image height in pixels. Default 1024.
+        num_steps: Number of diffusion steps. Default 4.
+        num_images: Number of images to generate. Default 1.
+    """
+    body = {
+        "prompt": prompt,
+        "width": width,
+        "height": height,
+        "numSteps": num_steps,
+        "numImages": num_images,
+    }
+    result = await _backend_post("/api/generate-image", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def retake_video(
+    video_path: str,
+    start_time: float,
+    duration: float,
+    prompt: str = "",
+    mode: str = "replace_audio_and_video",
+) -> str:
+    """Re-generate a portion of an existing video (retake / inpainting).
+
+    Args:
+        video_path: Local path to the source video file.
+        start_time: Start time in seconds for the retake region.
+        duration: Duration in seconds for the retake region.
+        prompt: Optional prompt to guide the retake.
+        mode: Retake mode. Default "replace_audio_and_video".
+    """
+    body = {
+        "video_path": video_path,
+        "start_time": start_time,
+        "duration": duration,
+        "prompt": prompt,
+        "mode": mode,
+    }
+    result = await _backend_post("/api/retake", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def suggest_gap_prompt(
+    before_prompt: str = "",
+    after_prompt: str = "",
+    gap_duration: float = 5,
+    mode: str = "t2v",
+    before_frame: str | None = None,
+    after_frame: str | None = None,
+    input_image: str | None = None,
+) -> str:
+    """Ask the AI to suggest a prompt for filling a gap in the timeline.
+
+    Args:
+        before_prompt: Prompt of the clip before the gap.
+        after_prompt: Prompt of the clip after the gap.
+        gap_duration: Duration of the gap in seconds.
+        mode: Generation mode. "t2v" (text-to-video) or others.
+        before_frame: Optional path to the last frame image before the gap.
+        after_frame: Optional path to the first frame image after the gap.
+        input_image: Optional input image path.
+    """
+    body: dict[str, Any] = {
+        "beforePrompt": before_prompt,
+        "afterPrompt": after_prompt,
+        "gapDuration": gap_duration,
+        "mode": mode,
+    }
+    if before_frame:
+        body["beforeFrame"] = before_frame
+    if after_frame:
+        body["afterFrame"] = after_frame
+    if input_image:
+        body["inputImage"] = input_image
+
+    result = await _backend_post("/api/suggest-gap-prompt", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def get_generation_progress() -> str:
+    """Get the current video generation progress (status, phase, step count)."""
+    result = await _backend_get("/api/generation/progress")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def cancel_generation() -> str:
+    """Cancel the currently running video generation."""
+    result = await _backend_post("/api/generate/cancel")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+# ===================================================================
+# ELECTRON BRIDGE TOOLS  (connect to project bridge on :8100)
+# ===================================================================
+
+@mcp.tool()
+async def list_projects() -> str:
+    """List all projects in LTX-Desktop. Each project contains assets and timelines.
+
+    Returns a JSON array of project objects with their IDs, names, and metadata.
+    """
+    result = await _bridge_get("/api/projects")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def get_project(project_id: str) -> str:
+    """Get a specific project's full data including assets, timelines, clips, and settings.
+
+    Args:
+        project_id: The project's unique ID (e.g. "project-1710000000000-abc123def").
+    """
+    result = await _bridge_get(f"/api/projects/{project_id}")
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def update_project(project_id: str, project_json: str) -> str:
+    """Update a project's data (assets, timelines, clips, etc.).
+
+    This tool edits the entire project JSON. Use get_project first to read
+    the current state, modify the needed fields, then send back the full
+    project JSON.
+
+    Args:
+        project_id: The project's unique ID.
+        project_json: The complete project JSON string. Must be valid JSON
+                      conforming to the Project type schema.
+    """
+    project_data = json.loads(project_json)
+    result = await _bridge_put(f"/api/projects/{project_id}", project_data)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def export_video(
+    project_id: str,
+    output_path: str,
+    width: int = 1920,
+    height: int = 1080,
+    fps: int = 24,
+    codec: str = "h264",
+    quality: int = 80,
+) -> str:
+    """Export the current timeline of a project as a video file using FFmpeg.
+
+    Args:
+        project_id: The project's ID to export from.
+        output_path: Local filesystem path for the output video (e.g. "/home/user/output.mp4").
+        width: Output video width. Default 1920.
+        height: Output video height. Default 1080.
+        fps: Output frame rate. Default 24.
+        codec: Video codec. Default "h264".
+        quality: Output quality (0-100). Default 80.
+    """
+    body = {
+        "projectId": project_id,
+        "outputPath": output_path,
+        "width": width,
+        "height": height,
+        "fps": fps,
+        "codec": codec,
+        "quality": quality,
+    }
+    result = await _bridge_post("/api/export", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+@mcp.tool()
+async def import_asset(
+    project_id: str,
+    file_path: str,
+) -> str:
+    """Import a local file (video, image, or audio) into a project's asset library.
+
+    The file will be copied into the project's assets directory and be
+    available for use on the timeline.
+
+    Args:
+        project_id: The project's ID to import into.
+        file_path: Absolute local path to the file to import.
+    """
+    body = {
+        "projectId": project_id,
+        "filePath": file_path,
+    }
+    result = await _bridge_post("/api/import-asset", body)
+    return json.dumps(result, ensure_ascii=False, indent=2)
+
+
+# ===================================================================
+# Entry point
+# ===================================================================
+if __name__ == "__main__":
+    logger.info("Starting LTX-Desktop MCP Server...")
+    logger.info("  Backend URL: %s", LTX_BACKEND)
+    logger.info("  Bridge URL:  %s", LTX_BRIDGE)
+    mcp.run()
diff --git a/tsconfig.node.json b/tsconfig.node.json
index d923a04d..951f9a10 100644
--- a/tsconfig.node.json
+++ b/tsconfig.node.json
@@ -2,13 +2,15 @@
   "compilerOptions": {
     "composite": true,
     "skipLibCheck": true,
+    "target": "ESNext",
     "module": "ESNext",
     "moduleResolution": "bundler",
     "allowSyntheticDefaultImports": true,
     "strict": true,
     "outDir": "dist-electron",
     "rootDir": "electron",
-    "esModuleInterop": true
+    "esModuleInterop": true,
+    "types": ["node"]
   },
   "include": ["electron/**/*.ts", "vite.config.ts"]
 }