# AI Agent Setup & Tooling
**YTAPI** provides machine-readable documentation and integration tools for AI coding assistants, LLM pipelines, and Model Context Protocol (MCP) clients.
***
Machine-Readable Endpoints [#machine-readable-endpoints]
For LLMs and agents that need to fetch documentation into context windows without parsing heavy HTML:
| Resource | URL | Description |
| :----------------- | :------------------------------------- | :---------------------------------------------------------------------------------- |
| **LLMs Index** | `https://docs.ytapi.dev/llms.txt` | Spec-compliant manifest of all documentation pages with short descriptions. |
| **Full LLMs Dump** | `https://docs.ytapi.dev/llms-full.txt` | Single concatenated Markdown document containing the entire platform documentation. |
| **Raw Page MDX** | `https://docs.ytapi.dev/{slug}.mdx` | Append `.mdx` to any documentation URL to retrieve raw, clean Markdown. |
In bash-based agents (e.g. Claude Code, Codex, Aider), fetch the entire API reference with:
```bash
curl -s https://docs.ytapi.dev/llms-full.txt
```
***
Model Context Protocol (MCP) [#model-context-protocol-mcp]
YTAPI runs a remote MCP server. There is nothing to install: point your client at the URL, then sign in to your YTAPI account when the client asks, or send your API key.
| | |
| :------------ | :------------------------------------------------------------------ |
| **URL** | `https://api.ytapi.dev/mcp` |
| **Transport** | Streamable HTTP |
| **Auth** | OAuth sign-in, or `Authorization: Bearer ` |
| **Billing** | Same credits as the REST API. Connecting and listing tools is free. |
No key yet? The agent can create the account for its user: see [Agent sign-up](/agent-signup).
Claude (claude.ai and Claude Desktop) [#claude-claudeai-and-claude-desktop]
Add YTAPI as a custom connector. No API key is needed, and free Claude plans can add one custom connector:
1. Open **Customize → Connectors** (also reachable from **Settings → Connectors**), click **+ Add**, then **Add custom connector**.
2. Enter `https://api.ytapi.dev/mcp` and click **Connect**.
3. Sign in to YTAPI, or create an account (200 free credits, no card), then click **Allow**.
Then ask Claude about any YouTube video, channel or playlist. Requests use your account's credits and appear under **Usage**. To disconnect Claude, revoke the "Claude" connected app under [API keys](https://ytapi.dev/app/api-keys).
Claude Code [#claude-code]
Install the YTAPI plugin, which adds the server and skills for summaries, channel digests and topic research:
```text
/plugin marketplace add ytapi/youtube-skills
/plugin install ytapi@ytapi
```
Then run `/mcp`, pick `plugin:ytapi:ytapi` and choose **Authenticate** to sign in. If Claude Code is logged in with a claude.ai account that already has the YTAPI connector, the connector shows up on its own.
To use an API key instead:
```bash
claude mcp add --transport http ytapi https://api.ytapi.dev/mcp \
--header "Authorization: Bearer YOUR_API_KEY"
```
Codex [#codex]
Export your key, then register the server:
```bash
export YTAPI_KEY=YOUR_API_KEY
codex mcp add ytapi --url https://api.ytapi.dev/mcp --bearer-token-env-var YTAPI_KEY
```
Or add it to `~/.codex/config.toml` yourself:
```toml
[mcp_servers.ytapi]
url = "https://api.ytapi.dev/mcp"
bearer_token_env_var = "YTAPI_KEY"
```
Cursor [#cursor]
Add to `~/.cursor/mcp.json` (all projects) or `.cursor/mcp.json` (one project):
```json
{
"mcpServers": {
"ytapi": {
"url": "https://api.ytapi.dev/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}
```
VS Code [#vs-code]
Add to `.vscode/mcp.json`. VS Code asks for the key once and stores it securely:
```json
{
"inputs": [
{
"type": "promptString",
"id": "ytapi-key",
"description": "YTAPI API key",
"password": true
}
],
"servers": {
"ytapi": {
"type": "http",
"url": "https://api.ytapi.dev/mcp",
"headers": { "Authorization": "Bearer ${input:ytapi-key}" }
}
}
}
```
Google Antigravity [#google-antigravity]
Add to `~/.gemini/config/mcp_config.json` (all workspaces) or `.agents/mcp_config.json` (one workspace). Antigravity reads `serverUrl`, not `url`:
```json
{
"mcpServers": {
"ytapi": {
"serverUrl": "https://api.ytapi.dev/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}
```
OpenClaw [#openclaw]
Add the server under `mcp.servers` in your OpenClaw config:
```json5
{
mcp: {
servers: {
ytapi: {
url: "https://api.ytapi.dev/mcp",
transport: "streamable-http",
enabled: true,
headers: { Authorization: "Bearer YOUR_API_KEY" }
}
}
}
}
```
Keep the key in OpenClaw's secret storage rather than in the file where you can.
Other stdio-only clients [#other-stdio-only-clients]
Bridge to the remote server with [`mcp-remote`](https://www.npmjs.com/package/mcp-remote) (requires Node.js):
```json
{
"mcpServers": {
"ytapi": {
"command": "npx",
"args": [
"-y", "mcp-remote", "https://api.ytapi.dev/mcp",
"--header", "Authorization:${AUTH_HEADER}"
],
"env": { "AUTH_HEADER": "Bearer YOUR_API_KEY" }
}
}
}
```
Tools [#tools]
| Tool | What it does | Credits |
| :-------------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :------------------------------------------------------------------------------- |
| `get_transcript` | Captions of a video as Markdown (a timestamp every few lines), plain text, or SRT or VTT (a start and end time on every caption line). Accepts a video ID or any YouTube URL, including Shorts. | 1 for the first page; later pages of the same transcript are free for 30 minutes |
| `get_video_info` | Title, channel, duration and the caption languages a video has. | 0 |
| `get_video_details` | Exact view, like and comment counts, publish time, description, chapters, keywords and related videos, starting with views per day and like and comment rates. | 1 |
| `get_video_heatmap` | YouTube's "most replayed" graph: the most replayed moment, up to five peaks and the graph in 20 parts, with timestamped links. Many videos have none; that error is free. | 1 |
| `get_comments` | A video's comments, top or newest first, about 20 per page, with likes and reply counts. Also reads a comment's replies. | 1 per page |
| `search_youtube` | Search videos, channels, playlists or Shorts. | 1 per page |
| `get_channel` | A channel's profile: subscriber and video counts, country, description and links. Accepts a handle, channel ID or URL. | 1 |
| `get_channel_videos` | A channel's uploads, newest, most popular or oldest first. Accepts a handle, channel ID or URL. | 1 per page |
| `get_playlist_videos` | A playlist's details and videos, page by page. | 1 per page |
Long transcripts come back in pages of 40,000 characters (set `max_chars` up to 100,000). When more remains, the result ends with the `offset` to pass on the next call. Errors, such as a video without captions, are returned as tool errors and cost nothing.
***
Code Mode & Function Calling [#code-mode--function-calling]
When orchestrating extraction pipelines via OpenAI, Anthropic, or Gemini tool calling, provide the JSON Schema directly:
```json
{
"name": "fetch_youtube_transcript",
"description": "Extract subtitles or transcripts from any YouTube video in structured Markdown or SRT.",
"parameters": {
"type": "object",
"properties": {
"video_id": {
"type": "string",
"description": "11-character YouTube video ID"
},
"format": {
"type": "string",
"enum": ["markdown", "text", "srt", "vtt", "word_timestamps"],
"default": "markdown"
}
},
"required": ["video_id"]
}
}
```
***
Platform Skills & System Prompts [#platform-skills--system-prompts]
Cursor Rules (.cursorrules) [#cursor-rules-cursorrules]
Add this prompt rule to your project to instruct Cursor on how to query YouTube transcripts:
```markdown
# YTAPI guidelines
When writing code that extracts YouTube subtitles or transcripts:
1. Always use `https://api.ytapi.dev/v1/transcripts` with `Authorization: Bearer $YT_API_KEY`.
2. For LLM summaries or context injection, specify `"format": "markdown"`.
3. For video subtitle synchronizing, specify `"format": "word_timestamps"` with `"word_level": true`.
4. Check `X-Cache` response headers (`HIT` or `MISS`) to measure latency.
5. Refer to complete documentation at `https://docs.ytapi.dev/llms.txt`.
```