About

A web search API for agents, priced so you can leave it running

Neurobird Search gives an AI agent live web results with the page content already extracted, plus clean markdown extraction of any URL. It is built and run in Europe by Neurobird. There are two tools, one price list, and a published set of things it does not do.

Why it exists

Search pricing assumed a human was reading the results

When a person runs a search, a few cents is invisible. When an agent runs forty searches to answer one question, and a hundred agents do it all day, the search bill stops being a rounding error and starts deciding what you are willing to let the agent try. Most search APIs are still priced for the first case. This one is priced for the second.

The other half is what comes back. A ranked list of links makes an agent fetch every page itself, which is a second round of latency and a second round of failures on the pages that block automated readers. Neurobird Search returns the relevant passages of each page in the same response, at no extra credit, because that is the part the model actually needs.

Verifiable numbers

Everything here is measured or published, nothing is projected

$2
per 1,000 basic searches. $4 deep, $6 with a grounded answer, $1 at a million credits a month
1,000
free credits every month, with no card and no signup form
~15s
measured median latency, basic. 22s standard, 28s advanced, about 5s to first ranked links when streaming
2
tools: web_search and extract_url, over the Model Context Protocol

Latency figures are measured medians on the current production deployment, not targets. Competitor prices quoted elsewhere on this site were read at each vendor's own site in August 2026 and are linked in place. Vendors change prices, so follow the link before you budget against a dated snapshot.

Honest limits

What this is not

No index of our own

Coverage and freshness are inherited from upstream search engines. If they cannot see a page, neither can we. A vendor running its own crawl can make guarantees we cannot.

Not the fastest

Median latency is measured in seconds, not milliseconds, because a call opens and reads real pages. If a person is waiting on the response, choose something faster.

No SLA, no SOC 2

There is no contractual uptime guarantee and no completed audit. If procurement needs either, this is not the right tool yet, and no amount of wanting the sale changes that.

Two endpoints, not a suite

Search and extract. No crawl, no sitemap mapping, no managed research agent. Vendors charging more are often charging for a wider product surface, which is a real difference.

Independently listed

Where to verify this exists without taking our word for it

Every claim on this page can be checked against a source that is not us. The server is published in the official Model Context Protocol registry, indexed by third party MCP directories, and the client facing metadata is public.

MCP Registry

Published as com.neurobird/search, with domain ownership proven by DNS on the neurobird.com apex.

Glama

glama.ai scans the live server and lists both tools.

Source and manifest

GitHub, MIT licensed, carrying the registry manifest and usage examples.

Machine readable

OpenAPI, llms.txt and a static MCP server card, all generated from the running service so they cannot drift from it.

Contact

Questions, invoices, or a limit you need raised

Write to [email protected]. If you are evaluating this against another vendor and want the comparison checked, say so and you will get a straight answer, including the cases where the other one wins.