What for: give AI neural web search via Exa — fresh sources with full page content and citations, built for real research.

What it gives the AI

The AI gets Exa neural search — search by meaning rather than keywords, returning fresh pages along with their full content (not just snippets). Perfect for research: the AI finds relevant sources, reads them and answers with quotes and links.

Example: "Find current rate-limiting practices for Fastify in 2026" — the AI uses Exa to find up-to-date articles and discussions, reads them in full and assembles an answer with links.

Setup (exact steps)

Two paths: the hosted URL (simplest) or a local npx server with an EXA_API_KEY (get one at dashboard.exa.ai/api-keys).

Claude Code

The hosted remote server (nothing to install locally):

claude mcp add --transport http exa https://mcp.exa.ai/mcp

Or locally with a key:

claude mcp add exa --env EXA_API_KEY=YOUR_KEY -- npx -y exa-mcp-server

Cursor / Claude Desktop

The local npx option with a key:

{
  "mcpServers": {
    "exa": {
      "command": "npx",
      "args": ["-y", "exa-mcp-server"],
      "env": {
        "EXA_API_KEY": "YOUR_KEY"
      }
    }
  }
}

The hosted alternative is the same block with "type": "http" and "url": "https://mcp.exa.ai/mcp" (no key needed).

Usage example

Say: "Give me an overview of how competitors handle SaaS onboarding — 5 examples with links" → the AI finds relevant pages through Exa, reads them and comes back with a structured summary and sources instead of making things up.

Security

Read-only against the web: the server searches and reads public pages, it changes nothing in your systems.
Token: the EXA_API_KEY is created in the Exa dashboard; don't commit it to a repository — keep it in a local scope or env var.
Supply chain: the exa-mcp-server package comes from exa-labs (★4.5k). The hosted URL removes the local-package question entirely.

Gotchas

Exa is a paid credit-based API: heavy research burns through your balance, so keep an eye on the dashboard.
Full page content is bulky — that hits both the MCP output limit and your credits; cap the number of results.
Quality depends on how you phrase the query: a vague prompt gives noisy results. Ask for specifics and a time range.

Exa Search MCP