MCP Exa Server
v3.2.1Exa Labs
Exa's official MCP server (exa-labs/exa-mcp-server v3.2.1, MIT) exposing neural/semantic web search, code search, crawling/content fetch, company and people research, and deep research tools. Hosted at https://mcp.exa.ai/mcp with an unauthenticated free tier (3 QPS, 150 calls/day) or an Exa API key for full access; also runs locally via npx exa-mcp-server. Read-only, agent-optimized results with source URLs; Instant Search (Feb 2026) delivers sub-150ms latency.
Trust Vector Analysis
Dimension Breakdown
๐Performance & Reliability+
Search quality review of Exa's neural/semantic retrieval approach
Latency review based on the Instant Search launch and published benchmarks
Content extraction tool review
Freshness assessment of index plus live-crawl fallback
Rate limiting behavior review across anonymous and keyed tiers
๐ก๏ธSecurity+
Credential handling review across local, hosted-keyed, and anonymous modes
Threat modeling of untrusted web content ingestion via search and fetch tools
Data handling review of query processing paths
Source attribution and verifiability testing
๐Privacy & Compliance+
Privacy policy review
Data retention terms review
Third-party sharing pathway analysis
๐๏ธTrust & Transparency+
Documentation completeness review
Attribution testing of result payloads
Source code and license review
Pricing documentation review
โ๏ธOperational Excellence+
Setup complexity assessment across hosted and local modes
Reliability assessment of the hosted service
Cost analysis of free and paid tiers
AI integration and agent-fit assessment
Repository activity and release-cadence analysis
- +Neural/semantic search purpose-built for AI agents, not repackaged keyword search
- +Instant Search (launched 2026-02-05) delivers sub-150ms latency for real-time agent loops
- +Unauthenticated hosted free tier (3 QPS / 150 calls/day) โ lowest-friction onboarding in its class
- +Broad tool set: web search, fetch/crawl, code search, company and people research, deep research
- +Read-only surface keeps worst-case blast radius low
- +MIT open-source server (~4.7k stars, v3.2.1) with configurable tool subsets to save context
- +All results carry source URLs for verification
- !Fetched web content flows into the LLM context โ indirect prompt-injection vector with no documented sanitization
- !Hosted endpoint accepts API keys via URL parameter, which can leak through logs and shared configs
- !Queries processed and retained by Exa's servers per its privacy policy; no self-hosted backend
- !Anonymous tier is tightly rate-limited (150 calls/day), requiring a key for real workloads
- !Search backend is proprietary despite the open-source MCP layer
- !Result quality varies with source websites, as with any web search
Use Case Ratings
code generation
Code search and doc lookup tools are strong for finding APIs, examples, and repos
customer support
Good for locating product docs and third-party resources
content creation
Excellent semantic research and fact-gathering with cited sources
data analysis
Useful for external data gathering, company and market research
research assistant
Outstanding: neural search, crawling, and deep-research tools built for agents
legal compliance
Useful for regulatory research; verify findings against primary sources
healthcare
Can surface medical literature; not suitable for clinical decisions
financial analysis
Strong company research tools; confirm figures from primary sources
education
Great for learning and semantic exploration of topics
creative writing
Semantic search surfaces inspiration and niche references well