Talk to any published book in the world right inside claude (or your other MCP clients).
- write mcp tool, test basic version
- if book too long, maintain in MCP cache, and paginate through it
- use gemini 1.5 for asking questions from book, feed that response to claude (due to GINORMOUS context length)
- spawn mini-RAG engine inside the server, ingestion, chunking, reranking. maintained through to end of app lifecycle
- integrate scihub as well. FREE THE WORLD'S KNOWLEDGEEE
- make sure you have uv installed
- install deps.
cd book-fetch &&
- add this to your
claude_desktop_config.json
{
"mcpServers": {
"book-fetcher": {
"command": "uv",
"args": [
"--directory",
"<PATH_TO_PARENT_DIR>/libgen-mcp",
"run",
"main.py"
]
}
}
}