BibiGPT

What the BibiGPT MCP server does

The BibiGPT MCP server gives an MCP-compatible agent access to BibiGPT’s media processing functions. Its primary use is turning a video, podcast, audio URL, or local media file into an AI-generated summary or transcript. The listed source types include YouTube, Bilibili, TikTok, podcasts, and other videos.

The repository is centered on the bibi skill and CLI. It also documents related skills for saved-library content, channel feeds, and visual analysis, but the MCP tools listed for this entry focus on watching media: summaries, chapters, subtitles, and long-running tasks.

How it works

A client can connect to the hosted MCP service at https://bibigpt.co/api/mcp, avoiding a local installation. The repository also supports a local desktop and CLI route. With that setup, the bibi command authenticates through BibiGPT and can process URLs or local files.

Short operations can run directly through summarize_video, while summarize_video_with_config accepts a custom prompt, model, and language. summarize_by_chapter produces a chapter-oriented result, and get_subtitle returns transcript or subtitle content with timestamps. For longer media, create_summary_task starts asynchronous processing and get_task_status checks its progress.

The BibiGPT MCP server is suitable for agents that need to answer questions about media, produce quick research notes, extract spoken content, or break a long recording into sections. The repository also describes workflows for batch processing, cross-source research, article rewriting, and exporting notes, although those workflows are separate from the six listed MCP tools.

Setup and configuration

The hosted route requires no local package installation according to the repository’s remote-MCP instructions. Add the MCP endpoint to a compatible client, then use the exposed tools. The repository specifically lists Claude, ChatGPT, Cursor, Claude Code, Codex, and OpenClaw among its supported usage environments; client-specific setup may differ.

For local use, install the BibiGPT desktop application or CLI. The documented installer supports macOS, Linux, and Windows installation paths. After installation, run bibi auth login; authentication opens a browser for OAuth and stores the token automatically. The repository states that API keys are not required for the Codex plugin installation.

The skill can be installed with npx skills add JimmyLv/bibigpt-skill. Codex users can instead add the repository as a plugin and install bibi from the marketplace. The remote service is updated server-side, while the skill files and local CLI have separate update procedures.

Tools and capabilities

• summarize_video: summarize a video or podcast URL.
• summarize_video_with_config: apply a custom prompt, model, and language.
• summarize_by_chapter: return a chapter-by-chapter summary.
• get_subtitle: retrieve subtitles or a transcript with timestamps.
• create_summary_task: start asynchronous processing for long videos.
• get_task_status: inspect the state of an asynchronous summary task.

Limitations and notes

The material does not specify quotas, pricing, retention, supported models, or guaranteed availability for individual source platforms. Results depend on the source media and the service’s ability to obtain usable audio or subtitles. The repository presents local-file processing through the CLI and agent skill; the listed MCP tool descriptions specifically refer to video or podcast URLs, so verify local-file support for the particular MCP deployment before relying on it.

The BibiGPT MCP server is distinct from the repository’s sibling skills. Library management, channel feeds, and frame or OCR analysis are documented as separate bibi-library, bibi-feed, and bibi-vision skills rather than as tools listed for this entry.