Search MCP servers

Web, local and vector search.

Search MCP servers give an assistant a live index of the web. This is the direct fix for the most familiar failure mode in an AI tool: an answer that was correct when the model was trained and has quietly stopped being correct since.

A search server returns results — titles, snippets, URLs — which the model then reasons over and cites. Paired with a fetch server, it becomes a full retrieval loop: search for the source, fetch the page, answer from what it actually says.

Most search servers in this category are API-backed, which means an API key and, past a free tier, a bill. That is the trade: an independent index that is not the model's memory, and does not go stale.

Search MCP servers in this directory

What people build with search MCP servers

Grounding answers in current sources

Ask about something that changed this month and get an answer built from pages that exist, with links you can check.

Researching a library before adopting it

Have the agent search for the current release, the open issues and what people are saying, rather than recalling the state of the project two years ago.

Checking a claim

Give the agent a statement and have it search for corroboration — useful precisely because it can come back and tell you it found none.

Finding local results

Some search servers expose a local index as well as a web one, which handles the 'near me' class of question a general index answers badly.

Choosing a search MCP server

  • Check the free tier against how you will actually use it. Agentic search issues several queries per question, so a monthly quota goes faster than it looks.
  • Prefer a server that returns snippets alongside URLs. Without them the agent has to fetch every result to find out whether it was relevant.
  • Look for freshness controls. Being able to restrict to the last week is what makes a search server useful for anything moving quickly.
  • Pair it with a fetch server. Search finds the page; fetch reads it. Neither is complete alone.

Frequently asked questions

Do I need an API key for a search MCP server?

Almost always, yes. These servers query a commercial search index, and the index authenticates you. Most offer a free tier that comfortably covers personal use.

Is this the same as the web search built into my AI client?

Similar in effect, different in control. A search MCP server uses an index you have chosen and can configure, works in any MCP client, and gives you the raw results rather than a vendor's summary of them.

How many searches will an agent make?

More than you would. A single question often becomes three or four queries as the agent narrows down. Budget accordingly when you pick a plan.

Install any of these

Every server on this page carries a one-line install command on its own listing. Paste it into your client config and restart — the search tools appear in your next session.

Or keep browsing: all MCP server categories · the full MCP server list · Agent Skills.