Do I need a Jina API key?

Not to start. Setting authentication to None gives you the limited anonymous tools, including reading a URL. A key with Bearer auth unlocks web search, reranking, embeddings, PDF extraction and higher limits.

Does Jina MCP run inside my bot?

No. It is a connected tool: your bot talks to Jina’s hosted MCP endpoint over HTTPS. Nothing runs in your container, and the URLs and text you ask about go to their service.

How is this different from the search provider my bot already has?

They complement each other. Search finds candidate pages; Jina’s reader turns one into clean text an agent can actually use, and reranking picks between candidates. For crawling a whole site, use a crawler such as Crawl4AI or Firecrawl instead.

← Home

Guide

Jina MCP

One connector, several jobs your bot otherwise does badly: searching the web, reading a page without the navigation and cookie banners, reranking results so the good one is first, and turning text into embeddings. Works on OpenClaw and Hermes.

Connected, Not Installed

Jina runs the MCP server; your bot talks to it over HTTPS at https://mcp.jina.ai/v1. Nothing runs in your container, so there is no disk cost and nothing to update — and, as with any connected tool, the URLs and text you ask about go to their service.

Connect It

Open the dashboard, go to Tools, and pick a running bot.
Find Jina MCP and connect it. The endpoint is filled in for you.
Create a Jina API key, leave the auth mode on Bearer, and paste it. That unlocks web search, reranking, embeddings, PDF extraction and higher limits.

You can also set authentication to None and use the limited anonymous tools, which include reading a URL. A reasonable way to try it before making an account, and enough if page-reading is all you want.

What Each Piece Is For

Reading a URL. Fetches a page as clean text instead of raw HTML. This is the one you will use most: it is the difference between an agent that summarises an article and one that chokes on a cookie wall.
Web search. Finding pages in the first place.
Reranking. Ordering a set of candidate results or passages by actual relevance — useful when you have twenty results and want the three that matter.
Embeddings. Turning text into vectors, for your own similarity search or retrieval.
PDF extraction. Pulling text out of a PDF at a URL without downloading it into the container first.

Where It Fits Beside Search You Already Have

Your bot already has a search provider — see web search. Jina MCP is not a replacement for that so much as a better set of hands for the reading half: search finds candidates, the reader turns one into usable text, and reranking picks between them.

If your need is crawling a whole site rather than reading a page, Crawl4AI and Firecrawl are the tools built for that.

Troubleshooting

Some tools are missing. You are probably on the anonymous mode. Add a Jina key with Bearer auth for the full set.
Rate limited. Anonymous and free-tier limits are real. A key raises them; heavy crawling wants a crawler, not this.
A page still comes back as noise. Some sites defeat every reader. Try the Live Browser, where you can log in or clear the interstitial yourself.

Try It

Connect it anonymously, give your bot a link to a long article, and ask for the three claims that matter. If that is useful, add a key and ask it to search first rather than being handed the link.