BrainTube MCP Server
Biến lịch sử xem YouTube, bài viết và nội dung web của bạn thành cơ sở kiến thức có thể tìm kiếm, truy cập từ Claude, Cursor và các công cụ AI khác qua giao thức MCP.
Tài liệu
Connect BrainTube to Any AI
BrainTube ships a native Model Context Protocol (MCP) server, so any MCP-capable AI client can read your transcripts, summaries, and highlights as first-class context.
What is MCP?
Model Context Protocol is an open standard from Anthropic that gives AI clients one uniform way to read external context. Instead of every tool building a bespoke integration, a client speaks MCP and the server exposes its data and tools through that shared interface.
BrainTube's MCP server is remote and hosted — there is nothing to install. You point a compliant client at the endpoint, authenticate, and your saved library becomes queryable from inside that client.
Supported clients
- Any MCP client — Native Model Context Protocol. Same endpoint, same config. If your assistant speaks MCP, it works — no integration from us required.
- Claude Desktop — Read summaries, search transcripts, and ask Claude to reason over your library.
- Cursor — Pull engineering talks and tutorials into your coding workflow as live context.
- ChatGPT — Connects in Developer mode, or via OpenAPI for custom GPTs. Both paths in testing.
- Perplexity — Add BrainTube as a custom remote connector so answers can draw on your own saved material alongside the live web.
- VS Code · GitHub Copilot — Native MCP support in Copilot agent mode — pull saved talks and docs in as context beside the codebase.
- Gemini CLI — Add the endpoint to your settings.json and OAuth is discovered automatically. The Gemini chat app does not take custom connectors yet.
- Codex CLI — Remote streamable-HTTP is supported — add the URL, then sign in with codex mcp login.
Setup
Endpoint: https://mcp.brain-tube.com/mcp
Claude.ai (one-click, OAuth): go to Settings → Connectors → Add custom connector. First type a name for it (e.g. BrainTube), then paste https://mcp.brain-tube.com/mcp in the URL field and sign in when prompted. No API key needed — Claude.ai connects via OAuth.
Cursor, Claude Desktop, CLI tools, and other MCP clients:
{
"mcpServers": {
"braintube": {
"type": "streamable-http",
"url": "https://mcp.brain-tube.com/mcp",
"headers": {
"X-BrainTube-Token": "bt_your_key_here"
}
}
}
}
- Using Claude.ai? Connect with the one-click OAuth flow above and skip to step 4. For Cursor, Claude Desktop, CLI tools and any other MCP client, create an API key at Settings → API Keys (https://app.brain-tube.com/settings).
- Paste the config above into your MCP client's config (or add the URL + X-BrainTube-Token header through the client's UI).
- Replace bt_your_key_here with your real key and restart the client.
- Try: "Summarize the last 5 videos I saved about pricing."
Available MCP tools
| Tool | Description |
|---|---|
search_knowledge | Full-text search over your personal BrainTube knowledge corpus. Searches across YouTube, Instagram, web, LinkedIn, GitHub, Twitter and more. Hybrid ranking (semantic + keyword, reciprocal-rank fusion); Bulgarian/Cyrillic queries are also searched in English. |
get_video | Get full details for a specific saved item including transcript, description, summary, key takeaways. Pass YouTube video ID or internal UUID. |
list_recent | List your most recently saved items across all source types. |
get_stats | Get your personal corpus statistics: total items saved, breakdown by source type, taint distribution. |
get_related | Find items semantically similar to a given item using vector similarity. |
search_by_source | Search your corpus filtered to a specific source type (youtube, web, notion, twitter, pdf, etc). |
search_by_date_range | Semantic search scoped to items saved between two dates. |
tag_item | Add or remove tags on a saved item. |
list_bookmarks | List saved bookmarks from your corpus. Filter by read/unread status. |
toggle_bookmark | Bookmark an item, remove a bookmark, or toggle its read/unread state. |
connect_readwise | Connect your Readwise account to BrainTube by saving your Readwise API token. |
sync_readwise | Import highlights from your Readwise library into your BrainTube corpus. |
ingest_notion_page | Ingest a single Notion page into your BrainTube corpus. |
add_note | Write a note or AI-generated synthesis back to a specific item in your corpus. |
get_recent_conversations | Retrieve your most recently saved Claude and ChatGPT conversations. |
get_expertise_profile | Analyse your corpus to build a knowledge expertise profile — expert, intermediate, surface, and blind spots by topic. |
get_session_brief | One-shot session bootstrap: expertise profile, last 5 AI conversations, and corpus stats in a single call. |
ingest_content | Ingest a single piece of content (note, article, document, etc.) into your BrainTube corpus. |
bulk_ingest | Ingest up to 50 items in a single call with dedup and embedding. |
random_resurface | Surface forgotten items from your corpus using weighted randomness. |
ingest_notion_database | Ingest all pages from a Notion database into your BrainTube corpus. |
set_notion_api_key | Save your Notion integration API key for ingest_notion_page and ingest_notion_database. |
backfill_embeddings | Generate vector embeddings for all items that are missing them. Required before semantic search. |
generate_api_key | Generate a personal API key for Obsidian Sync CLI or other integrations. Key shown once only. |
search_obsidian | Search your local Obsidian vault via the Obsidian Local REST API plugin. |
chat_with_brain | Ask a question to a public BrainTube Brain (curated knowledge base). Returns answer + source citations. |
list_brains | List all Brains you have created with slug, name, item count, and visibility. |
get_knowledge_graph | Build a knowledge graph around a specific item, showing how it connects to others via knowledge_edges. |
knowledge_health | Run a health check on your knowledge corpus: missing embeddings, stale items, gaps, overall score out of 100. |
get_knowledge_index | Get a topic-level index of your entire knowledge corpus grouped by primary topic. |
export_corpus | Export your entire BrainTube knowledge corpus as a ZIP of Markdown files. Returns a signed download URL. |
compile_knowledge | Generate concept articles from a topic cluster or Brain via the compile-knowledge edge function. |
get_concept_articles | Query compiled concept articles from your knowledge base, filter by cluster, brain, or search text. |
tag_cooccurrence | Find tags that frequently appear together across your corpus. |
entity_cooccurrence | Find named entities (people, orgs, tools) that frequently co-appear across your corpus. |
detect_gaps | Detect knowledge gaps: thin topics, entities without depth, stale high-value items. |
most_retrieved | Return the items you retrieve most often, ranked by retrieval_count. |
export_claude_md | Generate a CLAUDE.md-compatible knowledge context file for any Claude Code project. |
recompute_salience | Trigger a salience score recompute across your corpus. |
deep_search | Multi-hop knowledge search: semantic search + graph traversal from top results. |
retrieval_quality | Get a retrieval quality dashboard: search hit rates, zero-result queries, top terms. |
find_path | Find the shortest path between two items in your knowledge graph. |
compute_centrality | Recompute graph centrality scores across your corpus. |
get_edge_history | Get temporal history of knowledge edges between two specific items. |
Endpoint & authentication
Endpoint: https://mcp.brain-tube.com/mcp (MCP streamable-http)
Authentication accepts either an X-BrainTube-Token: bt_your_key_here header or an Authorization: Bearer bt_your_key_here header. Claude.ai can use OAuth instead of an API key.
Create and rotate keys at https://app.brain-tube.com/settings.
Frequently asked
What is MCP and why does BrainTube use it?
Model Context Protocol is an open standard from Anthropic that lets AI clients read external context through a unified interface. BrainTube ships a native MCP server so any compliant client — Claude Desktop, Cursor, Zed — can query your library without bespoke integrations.
Which AI clients work with BrainTube today?
Any client that speaks MCP. That's the point of building on an open standard instead of one integration at a time — when a new assistant ships MCP support, it reaches your library without us shipping anything. Claude Desktop and Cursor are the ones walked through step by step below; ChatGPT support is in testing. Anything speaking MCP streamable-http uses the same endpoint and the same config.
Do I need to be on a paid plan to use the MCP server?
No. MCP access is included on every plan, including Free. What differs by tier is the credit allowance: saving items, running searches and asking your brain all spend credits, and higher tiers get more of them. Free starts with 30 credits a month.
Where is my API key stored and how do I rotate it?
Keys are created in the BrainTube app under Settings → API Keys. They are shown once at creation, hashed at rest, and can be rotated or revoked at any time. Each key has its own audit log.
Can BrainTube read my private notes or only public videos?
BrainTube reads only what you save into your own library — public YouTube videos, articles you ingest, PDFs you upload. It never crawls third-party accounts on your behalf.
What happens to my data if I disconnect a client?
Disconnecting an MCP client removes its access. Your library, summaries, and highlights remain in BrainTube untouched. You can reconnect or export everything at any time.