search-rotation

Fetch one web page and return its main content as Markdown, rotating across extraction providers (Jina Reader, Firecrawl, Parallel, Tavily, Exa) with failover. Use it to read a URL the user or a previous search_web result already names; it does not discover links or crawl a site. Only http and https URLs are accepted. The result names the provider that succeeded, lists failed attempts and truncates content beyond 50,000 characters with a marker; if every provider fails, the tool returns an error listing each provider's reason.

Read this server's local engine configuration: enabled state, configured rotation order, per-capability access (keyless, configured key, or key/extra credentials still required), local usage counters and the quota source per engine. Use it before update_engine_config for valid engine ids, and as the machine-readable answer when you need to know which providers are active, how much quota is left or why a provider is skipped — the dashboard is the human-facing view of the same installation. The tool performs no search and no fetch: it is not a live health check, stored errors are historical observations, and a disabled engine is a configuration state, not proof that a key is missing. Provider balances can come from a cached lookup that the dashboard refreshes; unknown provider limits are reported as null rather than zero or unlimited.

Open the local dashboard of this server in a browser on the machine running the server, for a human: it is the only place to enter API keys and provider credentials, strict-free mode, port and access token, and it shows the interactive usage, quota and request-history views. Use it for those credential and network settings and when a person should look at the data; read configuration with get_engine_status and change engines with update_engine_config instead, because their results are machine-readable and this call only launches a browser. The browser opens on the server host, not on the client's machine, and a server started with --no-dashboard returns an error.

Search the web through a rotating pool of search providers (Tavily, Firecrawl, Parallel, Exa, Google PSE, DuckDuckGo) with automatic failover when one errors, is rate-limited or has no local quota left. Use it when the answer depends on current web pages and no URL is known yet; use fetch_url to read a page whose URL is already known. Returns numbered results (title, URL, snippet, publication date when the provider supplies one) and names the provider that answered plus every failed attempt, so failover stays visible. Time filters keep rotation inside the providers that implement them and fail explicitly instead of searching without the filter; if every eligible provider fails, the tool returns an error listing each provider's reason. The engine parameter only prefers a provider and leaves failover active.

Change the local engine configuration: enable or disable engines, set the search and fetch rotation order, and set monthly or daily call limits per engine. Use it when the user wants a provider switched on or off, reordered or capped; call get_engine_status first for valid ids, current order and access modes. The tool writes the shared configuration file of this installation, so the change applies to later searches of every client without a restart. API keys are not accepted here: an engine that still needs a key or extra credentials cannot be enabled until they are set in the dashboard (open_dashboard). Returns the effective per-engine configuration, or an error naming the offending ids — on error nothing is written.