# Exa MCP Server

> An open-source MCP server that connects AI assistants like Claude, Cursor, and VS Code to Exa's web search, crawling, and research agent capabilities.

Exa MCP Server is an open-source Model Context Protocol (MCP) server built by Exa Labs that gives AI assistants direct access to Exa's web search and crawling infrastructure. It is available on GitHub under the MIT License and can be connected to any MCP-compatible client via a hosted remote endpoint or a local npm package.

## What It Is

Exa MCP Server bridges AI coding assistants and chat agents to Exa's search API, enabling real-time web search, full-page content fetching, advanced filtered search, and multi-step agentic research — all from within the AI assistant's context window. The server implements the Model Context Protocol, a standard that lets AI tools call external services as structured tools. The core repository is written in TypeScript and has accumulated over 4,800 GitHub stars since its creation in November 2024.

## Available Tools and What They Do

The server exposes four tools, two enabled by default and two opt-in:

- **`web_search_exa`** — Fast web search returning clean, ready-to-use content; the default for everyday queries.
- **`web_fetch_exa`** — Fetches the full markdown content of one or more URLs.
- **`web_search_advanced_exa`** — Exposes the full Exa Search API with category filters (company, people, news, publication, financial report, personal site), domain restrictions, date ranges, highlights, summaries, geo-targeting, and subpage crawling.
- **`agent_run`** — Runs Exa Agent for multi-step research, list-building, data enrichment, and structured output with grounded citations; requires OAuth or an API key because agent runs are usage-based.

Additional tools are enabled by appending a `tools` query parameter to the hosted MCP URL (`https://mcp.exa.ai/mcp`).

## Client Compatibility and Setup Path

The server works with any MCP-compatible client. The documentation lists explicit setup instructions for Cursor, VS Code, Claude Desktop, Claude Code, Codex, OpenCode, Kiro, Windsurf, Zed, Gemini CLI, Warp, v0 by Vercel, Grok Build, LM Studio, Replit, and Roo Code. Most clients need only a single JSON config entry pointing at `https://mcp.exa.ai/mcp`. For clients that do not support remote MCP servers natively, an `mcp-remote` bridge or the `exa-mcp-server` npm package with a local API key can be used instead. Claude Desktop also supports Exa as a native Connector, requiring no config file at all.

## Agent Skills for Claude Code

The repository ships ready-to-use Claude Code skills — prompt templates that teach Claude how to use Exa search for specific tasks. Available skills cover company research, code search, people search, financial report search, research paper/publication search, and personal site search. Each skill includes token-isolation patterns (spawning Task agents to keep the main context clean) and category-specific filter guidance to avoid API errors.

## Open-Source Lineage and Adoption Signal

The repository is MIT-licensed and maintained by Exa Labs, the applied AI research company behind the Exa Search API. The about page states Exa powers search for over 400,000 developers and is used by teams including Cursor, Cognition (Devin), HubSpot, Monday.com, and OpenRouter — though these are vendor-published claims. The GitHub repository had 4,808 stars and 363 forks as of early August 2026, with active pushes as recently as August 1, 2026, indicating ongoing maintenance. The MCP server is listed on the Cursor marketplace and has a one-click VS Code install badge.

## Update: Exa Agent Tool Added to MCP

The most notable recent addition to the MCP server is the `agent_run` tool, which exposes Exa Agent — a multi-step research agent that runs asynchronously, supports structured output schemas, grounded citations, configurable effort levels (minimal, low, medium, high, xhigh, auto), and Exa Connect data source integrations. Long-running agent calls that exceed the call window return a run ID so the client can resume polling. This extends the MCP server from a simple search connector into a full agentic research interface accessible from any compatible AI assistant.

## Features
- Web search via MCP tool calls
- Full webpage content fetching as markdown
- Advanced search with category, domain, date, and text filters
- Exa Agent for multi-step agentic research
- Structured output with grounded citations
- Token-efficient highlights extraction
- Hosted remote MCP endpoint (no local install required)
- npm package for local deployment with API key
- OAuth and API key authentication
- Claude Code agent skills for company, code, people, financial, and publication search
- Configurable tool selection via URL parameters
- Support for 15+ MCP clients including Cursor, VS Code, Claude, Windsurf, Zed, Gemini CLI

## Integrations
Cursor, VS Code, Claude Desktop, Claude Code, Codex, OpenCode, Kiro, Windsurf, Zed, Gemini CLI, Warp, v0 by Vercel, Grok Build, LM Studio, Replit, Roo Code, LangChain, LlamaIndex, CrewAI, Pydantic AI, Browserbase, Snowflake, ElevenLabs, Slack, Google Sheets, Stripe

## Platforms
WINDOWS, MACOS, WEB, API, VSC_EXTENSION, DEVELOPER_SDK, CLI

## Pricing
Open Source, Free tier available

## Version
main

## Links
- Website: https://exa.ai/docs/reference/exa-mcp
- Documentation: https://exa.ai/docs/reference/exa-mcp
- Repository: https://github.com/exa-labs/exa-mcp-server
- EveryDev.ai: https://www.everydev.ai/tools/exa-mcp-server