OpenSERP

OpenSERP gives you live search results on localhost. There are no API keys to obtain and no per-search billing, and the engine list covers ground the paid APIs skip, which matters when the search is feeding an agent rather than a person.

It ships as a Go binary, a prebuilt Docker image or a go install, and serves an HTTP API on port 7000. Each engine has a dedicated endpoint and all of them return the same JSON schema, so switching sources does not mean rewriting a parser. Official SDKs exist for JavaScript and Python, along with an MCP server for AI agents and a community node for n8n.

Beyond plain queries, the server handles the work that usually sits downstream of a search call.

• Megasearch: one query hits several engines at once, and results are deduplicated by normalized URL into clusters that keep the per-engine ranks.
• URL extraction: the target pages come back as clean markdown in the same request.
• SERP features: AI summaries, answer boxes, people-also-ask entries and related searches are parsed out of the page.
• Filters and formats: language, date range, file type and site filters, image search, and JSON, markdown, text or NdJSON output.
• Operational extras: proxy support, caching and a resilient mode for long runs.

That suits two jobs in particular, grounding LLMs and agents in current results, and backing SEO rank tracking. A hosted version with the same API exists for teams that would rather not run the infrastructure themselves.