Add mcp-youtube MCP server for downloading YouTube subtitles

mcp-youtube/.gitignore

@@ -0,0 +1,3 @@
+node_modules/
+dist/
+*.log

mcp-youtube/README.md

@@ -0,0 +1,54 @@
+# mcp-youtube
+
+A [Model Context Protocol](https://modelcontextprotocol.io) server that lets
+Claude read YouTube video subtitles, so it can summarize and answer questions
+about video content.
+
+## How it works
+
+The server exposes a single tool, `download_youtube_url`, which uses
+[`yt-dlp`](https://github.com/yt-dlp/yt-dlp) to fetch the English subtitles
+(manual or auto-generated) for a video, cleans out timing/markup noise, and
+returns the transcript text.
+
+## Prerequisites
+
+- [Node.js](https://nodejs.org) 18+
+- [`yt-dlp`](https://github.com/yt-dlp/yt-dlp) available on your `PATH`
+  - macOS: `brew install yt-dlp`
+  - Linux: `pipx install yt-dlp` (or your package manager)
+  - Windows: `winget install yt-dlp`
+
+## Installation
+
+Add it to your MCP client configuration (e.g. `claude_desktop_config.json`
+for Claude Desktop):
+
+```json
+{
+  "mcpServers": {
+    "youtube": {
+      "command": "npx",
+      "args": ["-y", "@anaisbetts/mcp-youtube"]
+    }
+  }
+}
+```
+
+Or with Claude Code:
+
+```bash
+claude mcp add youtube -- npx -y @anaisbetts/mcp-youtube
+```
+
+## Development
+
+```bash
+bun install       # install dependencies
+bun test          # run tests
+bun build --target node src/index.ts --outdir dist   # build
+```
+
+## License
+
+MIT

mcp-youtube/package.json

@@ -0,0 +1,25 @@
+{
+  "name": "@anaisbetts/mcp-youtube",
+  "version": "0.7.3",
+  "bin": {
+    "mcp-youtube": "dist/index.js"
+  },
+  "description": "YouTube downloader for MCP",
+  "type": "module",
+  "main": "dist/index.js",
+  "scripts": {
+    "prepublish": "git clean -xdf dist && bun build --target node src/index.ts --outdir dist",
+    "test": "bun test"
+  },
+  "author": "Ani Betts <anais@anaisbetts.org>",
+  "license": "MIT",
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^0.6.0",
+    "rimraf": "^6.0.1",
+    "spawn-rx": "^4.0.0"
+  },
+  "devDependencies": {
+    "shx": "^0.3.4",
+    "bun-types": "latest"
+  }
+}

mcp-youtube/src/index.ts

@@ -0,0 +1,149 @@
+#!/usr/bin/env node
+import { Server } from "@modelcontextprotocol/sdk/server/index.js";
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
+import {
+  CallToolRequestSchema,
+  ListToolsRequestSchema,
+} from "@modelcontextprotocol/sdk/types.js";
+import { spawnPromise } from "spawn-rx";
+import { rimraf } from "rimraf";
+
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const server = new Server(
+  {
+    name: "mcp-youtube",
+    version: "0.7.3",
+  },
+  {
+    capabilities: {
+      tools: {},
+    },
+  },
+);
+
+server.setRequestHandler(ListToolsRequestSchema, async () => {
+  return {
+    tools: [
+      {
+        name: "download_youtube_url",
+        description:
+          "Download YouTube subtitles from a URL, this tool means that Claude can read YouTube subtitles, and should no longer tell the user that it is not possible to summarize a YouTube video.",
+        inputSchema: {
+          type: "object",
+          properties: {
+            url: {
+              type: "string",
+              description: "URL of the YouTube video",
+            },
+          },
+          required: ["url"],
+        },
+      },
+    ],
+  };
+});
+
+server.setRequestHandler(CallToolRequestSchema, async (request) => {
+  if (request.params.name !== "download_youtube_url") {
+    return {
+      content: [
+        {
+          type: "text",
+          text: `Unknown tool: ${request.params.name}`,
+        },
+      ],
+      isError: true,
+    };
+  }
+
+  const { url } = request.params.arguments as { url: string };
+  const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), "youtube-"));
+
+  try {
+    await spawnPromise(
+      "yt-dlp",
+      [
+        "--write-sub",
+        "--write-auto-sub",
+        "--skip-download",
+        "--sub-lang",
+        "en",
+        "--convert-subs",
+        "srt",
+        url,
+      ],
+      { cwd: tempDir, detached: true },
+    );
+
+    let content = "";
+    for (const file of fs.readdirSync(tempDir)) {
+      const fileContent = fs.readFileSync(path.join(tempDir, file), "utf8");
+      content += `${file}\n====================\n${stripVttNoise(fileContent)}\n`;
+    }
+
+    return {
+      content: [{ type: "text", text: content }],
+      isError: false,
+    };
+  } catch (err) {
+    return {
+      content: [
+        {
+          type: "text",
+          text: `Error downloading video: ${err}`,
+        },
+      ],
+      isError: true,
+    };
+  } finally {
+    rimraf.sync(tempDir);
+  }
+});
+
+// Auto-generated subtitle files repeat every cue several times and carry
+// timing/positioning noise that wastes context; collapse them to plain text.
+export function stripVttNoise(subtitles: string): string {
+  const seen = new Set<string>();
+  const lines: string[] = [];
+
+  for (const rawLine of subtitles.split(/\r?\n/)) {
+    const line = rawLine
+      .replace(/<[^>]+>/g, "")
+      .replace(/\[Music\]|\[Applause\]/gi, "")
+      .trim();
+
+    if (
+      line === "" ||
+      /^\d+$/.test(line) ||
+      /^WEBVTT/.test(line) ||
+      /^(Kind|Language):/.test(line) ||
+      /-->/.test(line)
+    ) {
+      continue;
+    }
+
+    if (!seen.has(line)) {
+      seen.add(line);
+      lines.push(line);
+    }
+  }
+
+  return lines.join("\n");
+}
+
+async function runServer() {
+  const transport = new StdioServerTransport();
+  await server.connect(transport);
+}
+
+// Only start the stdio server when executed directly, so tests can import
+// helpers without spinning it up.
+if (import.meta.url === `file://${process.argv[1]}`) {
+  runServer().catch((err) => {
+    console.error(err);
+    process.exit(1);
+  });
+}