Official Exa MCP for fast web search and crawling when the workflow is search-first rather than page-ops-first.
Where it wins
Neural/keyword hybrid search with semantic embeddings — unique capability
940K weekly downloads (270K npm + 670K PyPI) — massive real adoption
Exa Deep launch (March 4 2026) adds deep research tier ($12-15/1K)
Official provider support, SOC 2 Type II
Good fit for discovery and initial market mapping
Where to be skeptical
MCP benchmark: only 23% web search success — MCP implementation is immature
Exa Deep tiers are new (March 2026), limited production track record
Closed-source, cloud-only — no self-hosting option
Latency ~1,200ms for standard search (2x Brave)
Editorial verdict
#8 in product-business-development, #2 research tool. 4,050 stars, 9.3K npm/wk, 19.7K PulseMCP/wk (#56 global). Best for semantic discovery ('find companies like X'). exa-code for coding agents is a unique differentiator. AIMultiple benchmark tested extraction, not discovery — 23% undersells Exa's actual strength.
Videos
Reviews, tutorials, and comparisons from the community.
5 Claude Code MCP Servers You Need To Be Using
13 OpenClaw Skills You NEED To Install Right Now
Top 5 MCP Servers For Vibe Coding In 2026
Agents at Work 16: What LLMs are looking for with Thais Castello (Exa.ai)
Building a Smarter AI Agent with Neural RAG - Will Bryk, Exa.ai
Related
Product / Business Development
Semantic discovery — 'find companies like X,' market mapping, academic search
Search & News
Semantic search, similarity search, market mapping — strongest where traditional keyword search fails
Crawl4AI
93Free, open-source web scraping (Apache-2.0). 62K stars, 6,353 forks (nearly matches Firecrawl), actively maintained (v0.8.5, 2026-03-18), 384K weekly PyPI downloads. Best open-source alternative to Firecrawl.
SearXNG
88Privacy-first, self-hosted meta-search engine aggregating 70+ upstream engines. Zero cost, zero API keys, full data sovereignty.
Notion MCP Server
86Official Notion MCP server with hosted OAuth, token-optimized Markdown responses, and 22 tools for page/database/search operations.
MCP Atlassian
85Community-built MCP server for Jira and Confluence with 72 tools. Dual implementation alongside official Atlassian Rovo MCP (now GA).
Public evidence
Exa Fast claims 'most accurate search API sub-500ms.' Exa Deep adds agentic multi-search. New maxAgeHours for freshness control. Free MCP tier launched. If independently verified, closes latency gap with Brave.
Strongest MCP adoption in the search category. 5x more stars than Brave's MCP server (781). Suggests strong developer preference for Exa's MCP integration.
Scored 5/10. Praised setup simplicity and MIT license. Criticized lack of documentation on prompts, resources, and tools. No sampling support.
Step-by-step tutorial for using exa-mcp-server with Claude Desktop to access LinkedIn profiles via MCP. No sponsorship disclosure.
Tested Exa MCP with Claude Code for a day and found it 'quite good,' praising search quality and speed. Covers pricing, installation, and practical usage.
Listed as Top Pick. Released Nov 27, 2024. Ranked #52 globally with estimated 706K visitors. Independent aggregator positioning.
412pts is exceptional traction for a search API launch (top 0.1% of HN posts). Launched as The web as a database — neural search engine for AI.
Agent Score 14.39 (#3 of 8). Statistically tied with Brave and Firecrawl in the top tier. Independent benchmark validation.
Nvidia investment signals infrastructure-grade credibility. $700M valuation validates market position. Exa serves web search to thousands of companies including Cursor.
Exa scores 81% vs 71% on complex retrieval, runs 2-3x faster, and adds people, company, and code search that Tavily lacks.
23% MCP success rate vs Firecrawl's 83%. 15s avg vs Firecrawl's 7s. MCP-specific problem — Exa's core API performs better when consumed directly. Timeout and Cloudflare issues on hosted endpoint (GitHub #6878).
Two tiers: deep ($12/1K, 4-12s) and deep-reasoning ($15/1K, 12-50s). Structured outputs with field-level grounding. Claims to beat Perplexity Sonar and Parallel on HLE-Search, Deep Search QA, FRAMES — but no specific numbers published.
Exa scores 94.9% on SimpleQA (highest in comparison), 81% vs 71% on complex retrieval, runs 2-3x faster than Tavily. Independent accuracy edge.
Raw GitHub source
Constrained peek so you can sanity-check the source material without leaving the site.
Connect AI assistants to Exa's search capabilities: web search, code search, and company research.
Connect to Exa's hosted MCP server:
https://mcp.exa.ai/mcp
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"exa": {
"url": "https://mcp.exa.ai/mcp"
}
}
}
Add to .vscode/mcp.json:
{
"servers": {
"exa": {
"type": "http",
"url": "https://mcp.exa.ai/mcp"
}
}
}
claude mcp add --transport http exa https://mcp.exa.ai/mcp
Exa is available as a native Claude Connector — no config files or terminal commands needed.
That's it! Claude will now have access to Exa's search tools.
<details> <summary>Alternative: manual config</summary>Add to your config file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"exa": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.exa.ai/mcp"]
}
}
}
codex mcp add exa --url https://mcp.exa.ai/mcp
Add to your opencode.json:
{
"mcp": {
"exa": {
"type": "remote",
"url": "https://mcp.exa.ai/mcp",
"enabled": true
}
}
}
Open the MCP Store panel (from the "..." dropdown in the side panel), then add a custom server with:
https://mcp.exa.ai/mcp
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"exa": {
"serverUrl": "https://mcp.exa.ai/mcp"
}
}
}