## What problem does Search solve?

Agents that research the open web answer from what they could reach — and a
great many pages won't show their content to ordinary fetch tools. The tool
gets an error page, the agent summarises from nothing, and nobody notices.
Fixing that yourself means buying a search API and a scraping service, then
owning the glue between them. Search is the one integration that does the
whole job: it finds the sources, reads them in full — through a real browser
when the site demands one — remembers what it read, and keeps every read on
record so you can show what an answer was based on.

## Does it get past paywalls or logins?

No. Search reads pages a person could open in their own browser, and paces
itself per source so it's never a burden on the site being read. Paywalled and
login-gated sources aren't supported, and the bundled guidance tells agents
exactly that.

## Can I use it for bulk scraping?

No. Search is a research tool: it reads the pages your agents need, paced
per source at a human rate, with the queue there for politeness and
patience — not volume. Bulk scraping isn't supported or allowed, and
workspaces detected scraping in bulk are suspended and investigated. The
per-source pacing also means bulk use wouldn't be fast even if you tried.

## What clients does it work in?

Claude Code and Codex install from the GitHub marketplace
(`claude plugin marketplace add levirge/search`); Claude Desktop, OpenCode,
pi.dev and any other MCP client connect to `https://search.levirge.com/mcp`
over Streamable HTTP. A REST endpoint exists too.

## What do my agents actually call?

`search` returns ranked sources with titles, URLs and snippets. `fetch`
returns a page as clean markdown — or just the fields (JSON) or sections
(markdown) the agent asked for, and with `schema: true`, the source's shape
first. `fetch_async` and `fetch_results` queue a list politely and collect
the outcomes.

## How does it read sites that block other tools?

A full browser opens the page. The first time an ordinary read of a
source comes back empty, Search notes it and uses the browser for that source
from then on — no rules file, no per-source setup.

## How fast is it?

Under a second on a warm read; demanding sources that used to take ten to
twenty seconds open in about a second once the path is warm. Anything read
stays available for 24 hours, so repeat reads come from cache — no new fetch.

## What happens when a page can't be read?

The failure is recorded — URL, outcome, timestamp — and reported to the
agent. Search never hands back an empty page for an agent to summarise.

## Where does it run?

We host it. Your workspace is isolated from every other customer's, with
per-user access tokens you issue and revoke yourself. If research needs to
run somewhere specific, talk to us about deployment options.

## What does it cost?

Talk to us — [talk to us](/contact)
and we'll walk you through it on a call.