Skip to content

FAQ

Do I need SearXNG running locally?

You need a reachable SearXNG backend. Local (127.0.0.1:8890) is the default, but any URL you set with SEARXNG_MCP_BASE_URL works.

Can I run it without Docker?

Yes. Prefer the Claude Code plugin or uvx --from git+https://github.com/88plug/searxng-mcp searxng-mcp. Docker is optional for self-hosted HTTP.

Does it support rendered pages?

Yes. Rendered fetch is automatic for weak HTML pages and can be forced with rendered=True.

Is rendered fetch always enabled?

Yes for installs, conditional at runtime. searxng-mcp includes Playwright by default and uses local Chromium or Chrome when present. If not, the first rendered fetch downloads Playwright Chromium into the user cache.

Can this be exposed publicly?

Not by default.

It should be treated as a trusted-service deployment with auth or a reverse proxy if you expose streamable-http.

Why are there prompts if the AI client drives MCP?

The prompts are compatibility helpers for clients that can surface them.

The primary interface is still tools.

search_many widens coverage before deduping results.

It is the right choice when one query is too narrow or ambiguous.

Why use research instead of search_many?

Use research when you want merged search results plus fetched sources and citations in one workflow.

Why does the server keep raw payloads hidden?

So the model sees a compact answer while the full data remains available to capable clients.

What is the default transport?

stdio.

That is the best local default and the safest onboarding path.