What it does: Fetches the content of a single URL and returns it as readable markdown. HTML is converted to markdown so the result is easy to read and quote. When to use: Use this tool whenever you need to retrieve or analyze live webpage content (e.g. research, quoting, summarization, checking a link, comparing pages). Use cases (in order): - Recall: The LLM recalls or selects the relevant URL(s) from the user's question, its knowledge, context, or links the user provided. Then call this tool for each URL to get up-to-date content. One URL per call. - Crawling: When you need many pages or want to crawl multiple URLs, call this tool in a loop. One tool call per URL. Do not pass multiple URLs in one call. Iterate over the list and call once per URL. Rules and limits: - The URL must be a fully-formed, valid URL (http or https). - This tool is read-only. It does not perform actions that change state or have side effects. - Results may be live or cached; the tool does not guarantee real-time freshness. - Authentication is not supported. If the URL requires login or auth, an error will be returned. - If the server returns a non-200 status (e.g. 404, 500), the tool returns an error message instead of page content. - This fetch runs from an isolated environment. Hosts like localhost and private IPs may not work. - Binary content is not supported (e.g. media files, PDFs). Only text-like responses are returned. - Requests are bounded by a maximum timeout. Very slow or unresponsive URLs will return a timeout error. - For static assets or non-webpage URLs, use another tool if available.

What it does: Runs a web search for one query via Ollama's web search API. Returns a list of results. Each result has a title, a URL, and a short text snippet. Use this to discover links by topic; then use web_fetch to get full page content for any URL you choose. When to use: Use when the user wants to search the web by query, explore a topic, or find URLs about something. After you get results, pick one or more URLs from the list and call web_fetch with that URL to retrieve the full page as markdown. Workflow: Call this tool with a query (e.g. "Ollama new engine", "Deno router tutorial"). You get back up to 10 results (title, url, snippet per result). To read a full page, call web_fetch with the url from the result. One query per call; for multiple topics, call web_search once per query. Rules and limits: - Query is required. max_results is optional (default 5, max 10). - Requires OLLAMA_API_KEY in the environment or MCP server config (e.g. from your Ollama account). - Read-only, no side effects. Results may be cached by the search provider.