---
title: "Tavily MCP server: setup for Claude, Cursor and Claude Code, pricing and alternatives"
category: "guides"
url: https://keirolabs.cloud/blog/tavily-mcp-server
---

## The 60-second version

**Tavily runs a hosted MCP server at `https://mcp.tavily.com/mcp/` that gives Claude, Cursor and other MCP clients two core tools, `tavily-search` and `tavily-extract`, plus site mapping and crawling.** You connect it with one command in Claude Code (sign-in by OAuth) or by putting your API key in the URL. Tool calls bill against your normal Tavily credits.

- **Remote URL:** `https://mcp.tavily.com/mcp/?tavilyApiKey=<your-key>`, or `https://mcp.tavily.com/mcp` with OAuth.
- **Local alternative:** `npx -y tavily-mcp@latest` with `TAVILY_API_KEY` set.
- **Cost:** the same as the API: 1 credit per basic search, 2 per advanced, 1 credit per 5 extracted URLs. Free plan: 1,000 credits a month.
- **Defaults:** set default search parameters once with `DEFAULT_PARAMETERS` (env var locally, header remotely).

Setup steps and prices come from [Tavily's MCP docs](https://docs.tavily.com/documentation/mcp), the [tavily-mcp repository](https://github.com/tavily-ai/tavily-mcp) and [Tavily's credit docs](https://docs.tavily.com/documentation/api-credits), checked October 19, 2026.

## What does the Tavily MCP server do?

**It exposes Tavily's API as tools an agent can call without any glue code.** Once connected, the model decides when to search or read a page, and the MCP client runs the call.

| Tool | What it does | Typical use |
|---|---|---|
| `tavily-search` | Real-time web search with ranked results | "Find the latest pricing for X", news with a time window, domain-restricted search |
| `tavily-extract` | Pulls the main content from one or more URLs | Read the article or doc page the search found |
| Map | Builds a structured map of a site's pages | Understand a docs site before reading it |
| Crawl | Walks a site and extracts the pages | Ingest a whole documentation section |

Tavily's docs name the first two tools exactly; the repository describes the map and crawl capabilities without fixing their tool names, so check your client's tool list after connecting.

## How do you set up the Tavily MCP server?

### Claude Code

**One command, and Claude Code opens a browser window to sign in. No key in the URL:**

```bash
claude mcp add tavily-remote-mcp --transport http https://mcp.tavily.com/mcp/
```

With OAuth, Tavily uses the dashboard key named `mcp_auth_default` if you have one, otherwise your `default` key.

### Cursor

Add a remote server through `mcp-remote`, with your key in the URL (Cursor 0.45.6 or later):

```json
{
  "mcpServers": {
    "tavily-remote-mcp": {
      "command": "npx -y mcp-remote https://mcp.tavily.com/mcp/?tavilyApiKey=<your-api-key>",
      "env": {}
    }
  }
}
```

### Claude Desktop

Run the local package with your key in the environment:

```json
{
  "mcpServers": {
    "tavily-mcp": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": { "TAVILY_API_KEY": "tvly-YOUR_API_KEY" }
    }
  }
}
```

Tavily's docs pin older versions (`tavily-mcp@0.1.2` for Claude Desktop, `0.1.3` for Cursor) in their snippets, while the repository README uses `@latest`. Use `@latest` unless you need to freeze behavior.

### Setting default search parameters

**`DEFAULT_PARAMETERS` saves you from repeating options on every call.** Locally it is an environment variable; on the remote server it is a header with the same JSON:

```bash
export DEFAULT_PARAMETERS='{"search_depth": "basic", "max_results": 10, "include_images": false}'
```

Keep `search_depth` on `basic` by default. Advanced search costs twice as much, and an agent left to choose will often pick it.

## What does the Tavily MCP server cost?

**There is no separate MCP fee. Every tool call bills credits exactly like the API.**

| Call | Credits |
|---|---|
| Search, basic | 1 |
| Search, advanced | 2 |
| Extract, basic | 1 per 5 successful URLs |
| Extract, advanced | 2 per 5 successful URLs |
| Map | 1 per 10 pages (2 with instructions) |
| Crawl (10 pages) | 3 basic, 5 advanced |

| Plan | Credits / month | Price | Per credit |
|---|---|---|---|
| Researcher (free) | 1,000 | $0 | — |
| Project | 4,000 | $30 | $0.0075 |
| Bootstrap | 15,000 | $100 | $0.0067 |
| Startup | 38,000 | $220 | $0.0058 |
| Growth | 100,000 | $500 | $0.005 |
| Pay as you go | — | — | $0.008 |

**Agents spend credits faster than people expect.** A research question in Claude Code commonly triggers five to ten searches and a few extracts. At pay-as-you-go rates that is about $0.05 to $0.10 per question on basic depth, and double on advanced. The free 1,000 credits cover roughly 100 to 200 agent research sessions a month. More detail is in our [Tavily API guide](/blog/tavily-api-guide-2026) and [Tavily pricing breakdown](/blog/tavily-search-api-pricing-per-1000-requests-2026).

## Common problems

- **"No tools found" in Cursor.** Update to 0.45.6 or later, and check that `npx` is on the PATH Cursor sees.
- **Credits drain overnight.** An agent loop with advanced search on by default. Set `DEFAULT_PARAMETERS` with `search_depth: basic`, and cap results.
- **Wrong key used with OAuth.** Create a dashboard key named `mcp_auth_default`, so usage lands on the key you expect.
- **Key exposed in a shared config.** Use OAuth in Claude Code, or the `Authorization: Bearer` header on the remote server, instead of the key-in-URL form.

## Tavily MCP alternatives

**Every major web search API now ships an MCP server. They differ in what one tool call returns and what it costs.**

| MCP server | Search price per 1,000 | What a search call returns | Setup |
|---|---|---|---|
| Tavily | $8.00 (PAYG basic) | Results with content | Hosted, OAuth or key |
| Keirolabs | from $0.25 | Results; `web_research` adds full page markdown | Claude Code plugin |
| Brave | $5.00 | Results from Brave's own index | Local server, API key |
| Exa | $7.00 | Results with contents | Hosted |
| Parallel | $1.00 | Dense excerpts | Hosted |

**Keirolabs ships its MCP tools as a Claude Code plugin:**

```bash
/plugin marketplace add keirolabs-API/keiro-mcp
/plugin install keiro-mcp@keiro-mcp
```

That adds `web_search` (fast search), `web_research` (search plus full markdown of the top pages), `extract_url` and `answer` (a cited answer), plus free utility tools such as `check_credits`. Search costs 1 credit, research 3, and paid plans halve both. The free tier is 1,250 credits a month, so an agent that searches all day can run on it before you pay.

For a wider comparison of web search MCP servers, including setup for Claude Desktop, see [web search MCP servers](/blog/mcp-model-context-protocol-web-search-2026-guide) and [the best MCP servers for web search and scraping](/blog/best-mcp-servers-web-search-scraping-2026).