blopus — Python SDK

Official Python client for the Blopus web search + fetch API. Blopus is a cheap, fast web-search API backed by an owned index — built for bots and agents.

The SDK talks to exactly two data-plane endpoints: POST /v1/search and POST /v1/fetch on https://api.blopus.ai.

Install

pip install blopus

Optional framework adapters:

pip install "blopus[langchain]"     # blopus.langchain.BlopusSearch
pip install "blopus[llamaindex]"    # blopus.llamaindex.BlopusToolSpec
pip install "blopus[crewai]"        # blopus.crewai.BlopusSearchTool

News scoping (news_only)

Set news_only=True when the question is about events — what happened, who announced what, market reaction, election results, earnings news. It searches only sources with a real newsroom, from a dedicated index, so it is faster than an unscoped search and it drops vendor blogs, marketing pages and documentation that otherwise crowd news results.

# events → scope it, and pair with a freshness window
client.search("what did the Fed announce", freshness="pd", news_only=True)

# documentation / reference → leave it off
client.search("kubernetes ingress example")

# wants both the announcement AND the changelog → leave it off
client.search("what's new in Python 3.14")

Omitting it searches everything, so leaving it off is always the safe choice.

Authentication

Pass an API key (blp_live_...) directly, or set BLOPUS_API_KEY:

export BLOPUS_API_KEY="blp_live_xxx"

Quickstart

from blopus import Blopus

client = Blopus()  # reads BLOPUS_API_KEY

res = client.search("who won the game last night", count=5, freshness="pd", news_only=True)
for hit in res:
    print(hit.score, hit.title, hit.url)
print("remaining quota:", res.remaining_quota)

# Fetch indexed page content
doc = client.fetch("https://example.com/article")
print(doc.title, len(doc.content))

Async

import asyncio
from blopus import AsyncBlopus

async def main():
    async with AsyncBlopus() as client:
        res = await client.search("openai news", freshness="pw")
        docs = await client.fetch([r.url for r in res])  # batch fetch
        print(docs.count, "fetched,", len(docs.failed_results), "missing")

asyncio.run(main())

search(...)

client.search(
    query,
    count=10,                 # 1..50
    freshness="all",          # pd | pw | pm | p3m | p1y | all
    news_only=False,          # True = newsroom sources only. FASTER (dedicated index) and
                              # drops vendor blogs/docs. Use it for events; leave it off for
                              # documentation, tutorials, forums — or when you want both.
    include_domains=None,     # ["techcrunch.com", ...]
    exclude_domains=None,
    start_date=None,          # "YYYY-MM-DD" or epoch seconds
    end_date=None,
    language=None,            # "en", "pt", ...
    offset=0,                 # pagination, up to 200
    include_excerpt=False,    # opt in to longer excerpts
    excerpt_chars=None,       # up to 1200
)

Returns a SearchResponse (iterable over SearchResult):

res.query            # echoed query
res.results          # list[SearchResult]
res.count            # number of results returned
res.offset
res.more_results     # bool — more pages available
res.remaining_quota  # int — your remaining monthly units

# SearchResult fields:
# title, url, snippet, domain, site_name, favicon,
# published_at, age_seconds, language, score

Search always costs 1 credit, regardless of parameters or excerpt size.

fetch(url_or_urls)

# single URL -> FetchResult
doc = client.fetch("https://example.com/a")
doc.url, doc.canonical_url, doc.title, doc.content, doc.domain, doc.published_at, doc.language, doc.found

# list of URLs -> BatchFetchResponse
batch = client.fetch(["https://a.com", "https://b.com"])
batch.results          # list[FetchResult] that were found
batch.failed_results   # list[FetchFailure] (url, found=False)
batch.count            # number found (== credits billed)
batch.remaining_quota

Batches over the server cap of 50 URLs are automatically split into ≤50-URL calls, run sequentially with a small delay, and merged for you. Fetch bills per document found.

Errors

All errors subclass blopus.BlopusError:

Requests to 429/5xx/connection errors are retried with exponential backoff (honoring Retry-After); tune with Blopus(max_retries=...).

MCP

Blopus also exposes a hosted MCP server (search + fetch tools) at https://mcp.blopus.ai, using the same Bearer auth.

from blopus import mcp_config, print_mcp_config
print_mcp_config()        # prints the mcpServers JSON to paste into your MCP client
cfg = mcp_config()        # or get it as a dict

CLI

blopus search "who won the game" --count 5 --freshness pd --news-only
blopus search "openai" --include-domains techcrunch.com,theverge.com --json
blopus fetch https://example.com/a https://example.com/b
blopus mcp-config

License

MIT