Open in the interactive docs · markdown for agents: https://neuraldeep.tech/docs/book-api.md

The book "Agents and vibe-coding" · search and MCP

A practical course on working with AI agents — 42 chapters — exposed as search for agents. Your Hub key is required — the same sk- as for chat. The reader on the site is open to everyone without a key: neuraldeep.ru/learn/books/agenty-vajbkoding.

Search is hybrid: BM25 full-text plus vector search over giga-embeddings (370 chunks split on paragraph boundaries). They fail differently — full-text finds the exact word but is helpless when you ask in other words; vector search catches meaning but misses exact terms. Fusing them covers both gaps, so mode=hybrid is the default.

curl -G "https://neuraldeep.tech/api/v1/books/agenty-vajbkoding/search" \
  -H "Authorization: Bearer $YOUR_KEY" \
  --data-urlencode "q=how much do idle skills cost" \
  --data-urlencode "mode=hybrid" --data-urlencode "limit=5"

Endpoints

method what it does
GET /api/v1/books list of books
GET /api/v1/books/{slug}/toc table of contents
GET /api/v1/books/{slug}/chapter/{id} full chapter text (ch12, appА)
GET /api/v1/books/{slug}/search search: q, mode = hybrid|fts|vector, limit 1–20

MCP server — the same search as tools for your agent:

{
  "mcpServers": {
    "neuraldeep-book": {
      "type": "http",
      "url": "https://neuraldeep.tech/api/mcp/book"
    }
  }
}

Tools: book_search, book_chapter, book_toc. Transport is JSON-RPC over HTTP, the server keeps no state. Without a key — 401.

Limits. 60 requests per minute per key, counted per key owner rather than per address — extra keys share the primary key's budget. Model answers about the book stay in the reader: those need GPUs and follow the plan quota (free: 10 questions a day).

Reference sections (markdown for agents)