Search the web, research companies and people, find code, and read pages — directly from Grok Build.
This plugin connects Grok Build to Exa, a search engine built for AI agents. Exa indexes the full web plus specialized collections: 50M+ company pages, 1B+ people profiles, 100M+ research papers, financial reports, and news. The plugin bundles Exa's hosted MCP server — install it once, sign in to your Exa account in the browser, and it works automatically.
Exa is a search engine, not a scraper. Where other tools crawl pages you already know about, Exa finds the right pages in the first place — ranking by semantic relevance across specialized indexes.
- Category search — query 50M+ companies, 1B+ people profiles, 100M+ papers, news, and financial reports as structured verticals, not generic web results
- Semantic ranking — natural language queries ("AI startups in Detroit building autonomous vehicles") outperform keyword strings
- Content extraction —
web_fetch_exareads any page as clean markdown with JS rendering
Benchmarks (exa-labs/benchmarks): Exa leads on RAG groundedness (79.4% vs Brave 76.3%, Tavily 61.1%), people search recall (R@10: 94.5% vs Brave 77.9%), and code content extraction (ROUGE-L: 83.2 vs next-best 73.7).
In Grok Build, run /plugin and search for exa, then select it to install.
On first connection, Grok prompts you to sign in to your Exa account in the browser (OAuth). No API key to paste — new accounts get free credits at signup.
| Tool | What it does |
|---|---|
web_search_exa |
Semantic web search with inline category filters (category:company, category:people, category:research paper, ...) |
web_fetch_exa |
Read any webpage as clean markdown |
| Skill | What it does |
|---|---|
exa-search |
Research orchestrator: plans the work, fans out parallel subagent searches with category filters, compiles deduplicated, cited results |
Just ask naturally:
Find AI infrastructure startups in San Francisco and compare their funding
Search for the latest research papers on transformer efficiency
How do I use Python asyncio gather with exception handling?
Read https://react.dev/blog and summarize the latest post
- Exa Agent (June 2026) — frontier web research agents via API, combining model fusion with Exa's search for deep research, list-building, and entity enrichment
- Exa Deep revamp (March 2026) — faster, cheaper deep search with structured outputs and field-level grounding
- Exa Instant (February 2026) — sub-150ms search with neural quality
- Company Search (January 2026) — fine-tuned retrieval model for 50M+ companies
- People Search (January 2026) — 1B+ indexed profiles
MIT