machinelibrary

Python client for Machine Library: search full text across scholarly papers, books, patents, standards and Wikipedia (plus Reddit, Telegram and Discord), read bounded full text, and find passages inside long documents. Every result carries a canonical source_uri to cite.

Built for AI agents and RAG pipelines. Framework adapters: langchain-machinelibrary, llama-index-tools-machinelibrary. Prefer MCP? Connect https://mcp.machinelibrary.ai — see machinelibrary.ai/mcp.

Install

pip install machinelibrary
export MACHINELIBRARY_API_KEY=ml_...   # https://machinelibrary.ai/keys

Use

from machinelibrary import MachineLibrary

ml = MachineLibrary()
hits = ml.search("solid-state battery dendrite growth", limit=5)
for hit in hits:
    print(hit.title, hit.source_uri)
    print("  ", hit.text)

# Find the passages that support a claim, without downloading the whole text.
for passage in ml.search_in_document(hits[0].source_uri, "critical current density"):
    print(passage.text)

# Bounded full text (check `content_truncated`).
doc = ml.fetch_document("doi:10.1038/nature14539", max_tokens=8000)

AsyncMachineLibrary has the same methods as coroutines.

search(index="social") searches Reddit, Telegram and Discord. Dates for issued_after / issued_before are Unix seconds. Identifiers accept DOIs (doi:…, https://doi.org/…), pmid:, arxiv: and isbn: shorthands.

Billing and limits

Searches and document reads are charged per request to the account that owns the key; see pricing. InsufficientCreditError means the balance needs a top-up. Results are evidence to review, not proof that a claim is correct, and coverage is not exhaustive.

License

MIT