vsmutok/ytscrape v2.2.0

on GitHub

Added

• Transcript.to_srt() and to_vtt() for subtitle export, with UTF-8
  dump_srt() / dump_vtt() helpers accepting paths or text streams (#18).
• Optional MCP server: install with ytscrape[mcp] and run ytscrape-mcp
  (or python -m ytscrape.mcp_server) to expose search, video/channel
  metadata, channel uploads, comments (replies included) and transcripts as
  read-only Model Context Protocol tools. stdio by default, plus
  --transport streamable-http|sse for hosted setups; --language,
  --region and --timeout flags mirror the YouTube facade.
  All tools carry read-only tool annotations (readOnlyHint, openWorldHint,
  idempotentHint) so MCP clients can auto-approve them.
• MCP server proxy support: --proxy (both schemes) plus --http-proxy /
  --https-proxy for per-scheme routing; the standard HTTP_PROXY /
  HTTPS_PROXY / NO_PROXY environment variables are honoured as well.
• MCP responses are size-capped (truncated + hint). List tools and
  transcripts return has_more / next_cursor instead of an unbounded dump.
  max_results is capped at 50. filter, sort and detail
  (concise / full) are enums. get_video / get_channel default to a
  concise projection. Cursors are depth-capped at 1 000 items (each page
  re-reads the listing from the start), list_transcripts returns the same
  {items, count, has_more, next_cursor, truncated} envelope as the other
  list tools, and a transcript snippet cut mid-text says so in hint.