Alex·an·dria

The most efficient search for AI agents.
Not a Google MCP. A resolved, cited, deduped answer bundle.

For agents. Not for documents.

A web search returns a list of links. The agent reads, dedupes, and reconciles them. That is a tax, not an answer.

The Library returns one compact bundle: the entity is resolved, the graph is walked, aliases are deduped, belief differences are surfaced, and everything is cited. The agent reads once.

Capabilities

🔍 Answer bundle

Resolve a target, walk the typed graph, dedupe, group by role, surface belief differences. One call, one answer.

📚 All fields

arXiv (all categories), Wikipedia (all domains), Semantic Scholar (citation graph). Continuously ingested.

⚡ Instant grab

Library misses? Falls through to Wikipedia, arXiv, and web search in parallel. Always has an answer.

🔐 Provenance-gated

Every artifact has source refs, confidence, and verifier label. Trust is machine-readable.

💸 Pay per call

No signup, no API key, no subscription. 402 Payment Required + x402. Payment is the credential.

🧠 Hybrid resolution

Exact id → alias → title → sparse similarity. Dense embedder seam. Ambiguous targets return candidates.

Pricing

RoutePriceWhat it does
query.nodes / edges$0.002Flat artifact query
query.graph$0.003Graph query (nodes + edges)
retrieve_artifact$0.001Single artifact fetch
assemble.bundle$0.005Full answer bundle (anchor product)
catalog.describe$0.001Capability descriptor (likely free)
resolve_and_fetch.grab$0.003Live fallback (Wikipedia, arXiv, web)

Prices are sub-cent USDC via x402. Dev token bypasses all pricing.

Try it

Start the server and query the Library in 30 seconds:

# Install and start
git clone https://github.com/sebastbernal2/alexandria
cd alexandria
python -m venv .venv && source .venv/bin/activate
pip install -e .

# Start the MCP server (stdio)
python -m library.mcp.server

# Or start the HTTP server
python -m library.service.app

In another terminal:

# Discover what the Library holds
python -m library.surface.cli catalog --token alex-dev-token

# Get an answer bundle
python -m library.surface.cli assemble '{"target":"momentum"}' --token alex-dev-token

# Query by type
python -m library.surface.cli query '{"intent":"nodes","filters":{"types":["Paper"]},"output_shape":"summary"}' --token alex-dev-token

MCP client config

Add this to your claude_desktop_config.json or .cursor/mcp.json:

{
  "mcpServers": {
    "alexandria": {
      "command": "python",
      "args": ["-m", "library.mcp.server"],
      "env": {
        "ALEX_DEV_PAYMENT_TOKEN": "your-token-here"
      }
    }
  }
}

Differentiator

Every other knowledge tool for agents returns documents. The Library returns answers — resolved, deduped, cited, with conflicts surfaced. The agent does the reasoning. We do the retrieval.

GitHub MCP Manifest