# AIDataNordic/alexandria-mcp [Health: Active]

**Category:** 🎨 Art & Culture  
**Repository:** https://github.com/AIDataNordic/alexandria-mcp  
**GitHub Stars:** 1  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/aidatanordic-alexandria-mcp

## Description
Semantic search over 4.6 million text chunks from 20,000 classical philosophy and humanities works. Built for AI agents using FastMCP.

## Tools
Capabilities this server exposes over MCP:

- **search_texts** — Search the collection using natural language. Uses hybrid dense+sparse retrieval with cross-encoder reranking.
- **get_book_list** — List books in the collection filtered by author, subject or language.
- **ping** — Connectivity test.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "alexandria-mcp": {
    "url": "https://alexandria.aidatanorge.no/mcp"
  }
}
```

## Documentation

## What AIDataNordic/alexandria-mcp MCP server does

AIDataNordic/alexandria-mcp MCP server exposes a searchable collection of public-domain texts published before 1928. The corpus contains more than 4.6 million chunks from over 20,000 works sourced from Internet Archive collections, including material associated with Gutenberg, Americana, and European Libraries.

The collection covers philosophy and humanities subjects such as ethics, metaphysics, epistemology, logic, political philosophy, theology, Stoicism, Neoplatonism, existentialism, and the history of ideas. It includes authors such as Plato, Aristotle, Kant, Hegel, Nietzsche, Descartes, Spinoza, Aquinas, Augustine, Marx, and others. Supported language codes include English, German, Latin, French, Italian, Greek, and Russian.

## How it works

The primary `search_texts` tool accepts a natural-language query and optional author and language filters. Results are limited to five by default and twenty at most. Each result can include the work title, creator, date, language, subject, retrieved text chunk, rerank score, and vector score.

Search uses a hybrid approach: dense embeddings and sparse BM25 retrieval first identify candidates, then a cross-encoder reranks them. The architecture stores vectors in a Qdrant collection named `alexandria`. The listed embedding model is multilingual E5 large, and the reranking stage uses an mMiniLM model.

`get_book_list` provides catalog-level discovery. It accepts optional author, subject, and language filters, with a default limit of 20 and a maximum of 100 books. A `ping` tool is available for connectivity checks. The README also lists prompts for analyzing a philosopher, exploring a topic across thinkers, and comparing two philosophers on a topic.

## Setup and configuration

The hosted endpoint is available at `https://alexandria.aidatanorge.no/mcp` over HTTP. A client can register that URL as an HTTP MCP server, or call the endpoint using JSON-RPC with the `tools/call` method. The documented Claude command uses the HTTP transport.

Self-hosting requires a running Qdrant instance with the `alexandria` collection already populated. The documented local process installs dependencies from `requirements.txt` and starts `alexandria_mcp_server.py`; the local MCP endpoint is then served at port 8005 by default. Optional configuration variables are `QDRANT_HOST`, `QDRANT_PORT`, and `MCP_PORT`, which default to `localhost`, `6333`, and `8005` respectively.

## Tools and capabilities

- Search text passages with natural-language queries.
- Narrow searches by author and language.
- Browse books by author, subject, and language.
- Check endpoint connectivity with `ping`.
- Retrieve ranking and source metadata alongside matching chunks.
- Use the listed philosopher-analysis, topic-exploration, and comparison prompts where the connected client supports MCP prompts.

## Limitations and notes

AIDataNordic/alexandria-mcp MCP server focuses on a fixed public-domain corpus rather than general web search or current publications. The README describes the included works as pre-1928 texts, so it is not a suitable source for recent scholarship. Search limits are capped at 20 results per request, and book-list results at 100.

Self-hosting is not a standalone empty-index deployment: Qdrant must be running and populated with the expected collection. The provided material does not document authentication, API keys, persistence setup, or procedures for importing additional books. The project is released under the MIT license.

_Full upstream README: https://allmcps.com/mcp/aidatanordic-alexandria-mcp/readme_