MCP server + CLI that gives AI coding agents web search capabilities for resolving errors.
AI coding agents (Claude Code, Cursor, Copilot, etc.) get stuck on errors they can't resolve — version mismatches, obscure library bugs, cryptic messages. The fix is what any developer would do: google it. just-google-it gives agents that ability.
# Set up in your project
npx just-google-it init
# Set your search API key
export BRAVE_API_KEY=your-key-here
# Test it works
npx just-google-it testGet a free Brave Search API key at brave.com/search/api ($5/month free credits, ~1000 queries).
Three tools are exposed to your AI agent via MCP:
| Tool | Purpose | Example |
|---|---|---|
search_errors |
Search for solutions to error messages | Agent hits ModuleNotFoundError, searches for fixes |
search_query |
General web search | "how to configure webpack 5 with TypeScript" |
search_docs |
Search library documentation | React useEffect cleanup docs |
With hooks enabled, just-google-it automatically searches after consecutive similar errors:
- Agent runs a command → fails
- Agent tries again → fails with similar error
- Hook fires → searches the web → injects solutions into context
No manual intervention needed.
npx just-google-it initThis will:
- Detect which AI agents you use (Claude Code, Cursor, Copilot, Windsurf)
- Configure MCP server connections
- Add search instructions to agent config files (CLAUDE.md, .cursorrules, etc.)
- Set up auto-trigger hooks (Claude Code only)
Add to your MCP configuration:
Claude Code (.claude/settings.local.json):
{
"mcpServers": {
"just-google-it": {
"command": "npx",
"args": ["-y", "just-google-it"],
"type": "stdio"
}
}
}Cursor (.cursor/mcp.json):
{
"mcpServers": {
"just-google-it": {
"command": "npx",
"args": ["-y", "just-google-it"]
}
}
}Create a .just-google-it.yml in your project root:
search:
backend: brave # brave | google | searxng
api_key_env: BRAVE_API_KEY
max_results: 5
timeout: 10000
domains:
preferred:
- stackoverflow.com
- github.com
blocked: []
errors:
auto_trigger: false
auto_trigger_threshold: 2
output:
include_code_snippets: true
include_urls: true
max_summary_length: 500| Backend | Setup | Cost |
|---|---|---|
| Brave Search (default) | Set BRAVE_API_KEY |
Free tier: ~1000 queries/month |
| Google Custom Search | Set GOOGLE_API_KEY + GOOGLE_CSE_ID |
Free tier: 100 queries/day |
| SearXNG | Self-hosted instance | Free (self-hosted) |
# Set up for your project
npx just-google-it init
# Test configuration
npx just-google-it test
# Search from command line
npx just-google-it search "webpack module federation error"- Error parsing: Strips noise (file paths, timestamps, ANSI codes) from error messages to create effective search queries
- Web search: Queries your configured search backend (Brave, Google, SearXNG)
- Solution extraction: Fetches top result pages and extracts solutions using heuristics:
- Stack Overflow: accepted answers + code blocks
- GitHub Issues: resolution comments
- General pages: main content + code blocks
- No extra LLM: The calling agent IS an LLM. Adding another would be slow, expensive, and redundant. Solutions are extracted with HTML parsing.
MIT