AI-driven YouTube Video Summarizer
BibiGPTAI.toWatch

https://bibigpt.co/api/mcp

Video summarizer agent skill · Claude Code · Codex · ChatGPT

The app is optional.
Your agent can watch the video.

Give Claude, ChatGPT, Cursor, terminals, and your own backend the same audiovisual pipeline. Hosted MCP talks to the API. Desktop and terminal use the local bibi CLI — not a second product.

Connect the MCPGet CLIOpen API docs

Works with

MCP, CLI, or API? Pick your interface

All three wrap the same audiovisual pipeline. Choose based on where the agent runs.

Tool-calling agents

BibiGPT MCP

Claude, ChatGPT, Cursor, VS Code, Grok Bot, Grok Build, and any MCP client. One-click OAuth, or a Bearer token. Hosted MCP talks to the API.

https://bibigpt.co/api/mcp

Desktop, terminal, CI

BibiGPT CLI

The local bibi binary from the desktop app — not a second product. Always pass --json. Without desktop, npx skills uses the same pipeline.

bibi summarize "<url>" --json

Backends and bulk jobs

Public API

REST for the same audiovisual pipeline. The OpenAPI 3 spec is tool-calling ready.

https://bibigpt.co/api/openapi.json

Connect MCP (including any client)

Grok and any unlisted host: open Grok or Any client. URL, OAuth / Bearer, Grok CLI / config.toml, Skill, and a usage prompt are all copyable. Named hosts still have their own tabs.

Remote MCP server URL
https://bibigpt.co/api/mcp
  1. If your host is missing, use this tab. Paste the URL or JSON into any MCP client. Grok Bot is /agent/grok-bot; Grok Build is /agent/grok-build — do not mix them.
  2. OAuth-capable hosts open a browser on first use. New accounts register in that flow — you do not need to sign up on the website first. Discovery: https://bibigpt.co/.well-known/oauth-protected-resource
  3. Headless hosts: copy a token from https://bibigpt.co/user/integration and send Authorization: Bearer.
  4. Coding agents can install the Skill with npx skills. Then paste the usage prompt.
Remote MCP server URL
https://bibigpt.co/api/mcp
mcp.json (OAuth)
{
  "mcpServers": {
    "bibigpt": {
      "url": "https://bibigpt.co/api/mcp",
      "type": "http"
    }
  }
}
mcp.json (Bearer token)
{
  "mcpServers": {
    "bibigpt": {
      "url": "https://bibigpt.co/api/mcp",
      "headers": {
        "Authorization": "Bearer $BIBI_API_TOKEN"
      }
    }
  }
}
Skill (any coding agent)
npx skills add JimmyLv/bibigpt-skill
Usage prompt
用 BibiGPT 总结这个视频：https://www.bilibili.com/video/BV1Sk4y1x7r2
Use BibiGPT to summarize this video: https://www.bilibili.com/video/BV1Sk4y1x7r2

Four skills, one CLI

Multiple skills, the same bibi binary, the same account. npx skills add JimmyLv/bibigpt-skill still installs bibi.

  • bibi

    Video summarizer agent skill: watch video — summarize, transcript, chapters. Not for saved-library search or channel feeds.

    npx skills add JimmyLv/bibigpt-skill --skill bibi

  • bibi-library

    Saved videos, notes, collections, in-library search.

    npx skills add JimmyLv/bibigpt-skill --skill bibi-library

  • bibi-feed

    Subscribe to channels, pull latest, mark feed seen.

    npx skills add JimmyLv/bibigpt-skill --skill bibi-feed

  • bibi-vision

    Frames, OCR, slides, mind map. Not digital humans or lip-sync.

    npx skills add JimmyLv/bibigpt-skill --skill bibi-vision

Agent-readable quick reference

A copy-paste block for crawlers and agents. Humans can copy it too.

quick reference
# BibiGPT MCP quick reference
what: video summarizer agent skill for Claude Code, Codex, ChatGPT, Cursor, Grok Bot, Grok Build, OpenClaw
endpoint: https://bibigpt.co/api/mcp
transport: Streamable HTTP
named grok hosts: Grok Bot → https://bibigpt.co/agent/grok-bot (custom MCP in chat). Grok Build → https://bibigpt.co/agent/grok-build (grok mcp add). Do not mix them.
any client: paste https://bibigpt.co/api/mcp into any MCP host we have not listed. Do not invent a host page.
auth (hosted clients): OAuth 2.1 with PKCE in-flow — new accounts register during login; discovery via https://bibigpt.co/.well-known/oauth-protected-resource
auth (header clients / CLI / API): Authorization Bearer token from https://bibigpt.co/user/integration
primary flow: summarize_video → optional get_subtitle → after save: list_saved_videos / search_saved_videos / get_saved_video → optional get_latest_feed / subscribe_channel
same pipeline: MCP (hosted, talks to the API) · desktop/terminal CLI (local `bibi`, not a second product) · Public API
cli: bibi summarize "<url>" --json
skill: npx skills add JimmyLv/bibigpt-skill
skills: bibi (Video summarizer agent skill: watch video — summarize, transcript, chapters. Not for saved-library search or channel feeds.) | bibi-library (Saved videos, notes, collections, in-library search.) | bibi-feed (Subscribe to channels, pull latest, mark feed seen.) | bibi-vision (Frames, OCR, slides, mind map. Not digital humans or lip-sync.)
docs: https://bibigpt.co/llms.txt, https://bibigpt.co/developers, https://bibigpt.co/agent, https://github.com/JimmyLv/bibigpt-skill, https://bibigpt.co/mcp
openapi: https://bibigpt.co/api/openapi.json
quota: Plus 100/day, Pro 300/day on the agent-skill channel; extra calls use API balance; upgrade /shop
tools: summarize_video, summarize_video_with_config, summarize_by_chapter, get_subtitle, create_summary_task, get_task_status, list_saved_videos, get_saved_video, search_saved_videos, list_notes, get_note, update_note, generate_summary_by_prompt, list_channels, subscribe_channel, unsubscribe_channel, get_channel_videos, get_latest_feed, get_channel_health, mark_feed_seen, list_collections, get_collection, create_collection, add_to_collection, get_collection_chat_history, generate_video_mindmap, extract_video_visuals, get_notion_status, export_to_notion, get_account_info

Agent workflow

Summarize → optional transcript → list/search/get after save → optional feed/subscribe.

  1. 1

    summarize_video

    Call summarize_video on a public audiovisual URL (> 30 min: create_summary_task then get_task_status).

  2. 2

    get_subtitle

    Call get_subtitle only when the transcript is needed — do not fetch it by default.

  3. 3

    list_saved_videos

    After it is saved, list / search / get with list_saved_videos, search_saved_videos, get_saved_video.

  4. 4

    get_latest_feed

    To follow a channel, subscribe_channel then poll get_latest_feed.

MCP tools exposed

/developers currently lists the summarize tools only. This page lists every enabled tool, grouped, with stable names.

Understand

  • summarize_videoDefault summary for a single < 30 min video / podcast (timestamped)
  • summarize_video_with_configCustom prompt / language / model
  • summarize_by_chapterChapter breakdown for long videos
  • get_subtitleTranscript only, fastest path
  • create_summary_taskAsync job for > 30 min; returns task_id
  • get_task_statusPoll status / fetch async result

Library

  • list_saved_videosList saved / summarized videos
  • get_saved_videoGet one video with summary, chapters, subtitles, note
  • search_saved_videosSearch across the saved library
  • list_notesList notes
  • get_noteGet a note
  • update_noteUpdate a note
  • generate_summary_by_promptRe-summarize a saved item with a custom prompt

Feed / channels

  • list_channelsList subscribed channels
  • subscribe_channelSubscribe to a channel
  • unsubscribe_channelUnsubscribe from a channel
  • get_channel_videosList videos on a channel
  • get_latest_feedLatest feed since last seen
  • get_channel_healthChannel fetch health
  • mark_feed_seenMark feed as seen

Collections / export / vision / account

  • list_collectionsList collections
  • get_collectionGet a collection
  • create_collectionCreate a collection
  • add_to_collectionAdd an item to a collection
  • get_collection_chat_historyCollection chat history
  • generate_video_mindmapGenerate a mind map
  • extract_video_visualsExtract visuals (PPT / OCR / slides)
  • get_notion_statusNotion connection status
  • export_to_notionExport a note to Notion
  • get_account_infoAccount and plan info

Run BibiGPT from any terminal

bibi ships with the desktop app. Pass --json so agents can parse stdout. Without desktop, install the Skill — same pipeline.

Install desktop / CLISkill GitHub
Without the desktop app
npx skills add JimmyLv/bibigpt-skill
Real CLI commands
bibi summarize "https://www.bilibili.com/video/BV1Sk4y1x7r2" --json
bibi library list --json
bibi library search --keyword "AI agents" --json
bibi call library.list --json

The same pipeline over REST

Spec: OpenAPI 3. Token from /user/integration. This is a real summarize call, not a sketch.

OpenAPI
https://bibigpt.co/api/openapi.json
POST /api/v1/summarize
curl -s -X POST "https://bibigpt.co/api/v1/summarize" \
  -H "Authorization: Bearer $BIBI_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://www.bilibili.com/video/BV1Sk4y1x7r2"}'

Frequently asked questions

What if my assistant is not listed?

Use the Any client tab. Paste https://bibigpt.co/api/mcp into any MCP host. OAuth clients sign in (and can register) in the same browser flow — no pre-registration. Headless clients copy a token from https://bibigpt.co/user/integration and send Authorization: Bearer. Skill install: npx skills add JimmyLv/bibigpt-skill.

What is the BibiGPT MCP?

The BibiGPT MCP is a Streamable HTTP Model Context Protocol server at https://bibigpt.co/api/mcp. Agents call summarization, transcript, library, and feed tools through it. Auth is OAuth2 plus Bearer.

Do I need an API key?

OAuth-capable clients (Claude, ChatGPT, Cursor, VS Code, Grok Bot) sign in with a BibiGPT account — new accounts register in the same flow, no key to copy. Header-configurable clients, the desktop CLI, and direct API calls use a token from /user/integration.

CLI vs MCP?

Same audiovisual pipeline, different interface. Hosted / cloud agents use MCP (which talks to the API). Local terminals, CI, and the desktop app use the bibi CLI. The Skill (npx skills add) is for coding agents without the desktop app.

How does quota work?

On the agent-skill channel, Plus gets 100 free calls/day and Pro 300. Extra calls use API balance. Web-app membership quota is separate. Upgrade: /shop

Upgrade at /shop. Plus 100/day, Pro 300/day.