Provides AI agents with web search capabilities via the Brave Search API. One of the lowest-risk MCP servers as it only performs read operations against public web content.
Security Rating
92 / 100
Vendor Type
Official (Anthropic)
Risk Category
Low Risk
Exposed Tools & Function Calls
- brave_web_search
- brave_local_search
Required Permission Scopes
- brave_search:read
Vulnerability & Risk Analysis
Indirect prompt injection via search results (Medium)
Malicious web pages indexed by Brave Search can include hidden text instructing the agent to take unintended actions after reading search snippets.
Fix: Parse search results as data, not as instructions. Implement strict instruction hierarchy where search results are treated as untrusted input.