One tool instead of many search tabs
Ask for web, shopping, company, finance, or site discovery results through one interface. Your runtime does not need custom logic per search engine.
Give your agents one search interface that returns clean, typed results. Guardian handles engine fan-out, normalization, and failures so your model can keep working.
Guardian replaces page scraping with stable response shapes. That cuts prompt bloat, reduces brittle browser flows, and keeps retrieval easy to observe.
Ask for web, shopping, company, finance, or site discovery results through one interface. Your runtime does not need custom logic per search engine.
Responses stay concise while preserving the fields an agent needs for ranking, citations, and downstream steps.
Build flows that depend on consistent JSON, not a rendered page that changes layout every other week.
Use MCP tools such as search_web, search_shopping, discover_sites, and finance_quote.
Guardian fans out queries, deduplicates overlapping results, and reports which engines succeeded or failed.
When a browser is still the right tool, Agent Mode keeps search controls and URLs predictable for automation.
Inspect elapsed time, provider information, paging, and diagnostics without scraping hidden UI state.
Use the same search service from direct HTTP clients, orchestrators, or agent frameworks.
The canonical transport stays at https://mcp.guardianbrowser.sh/v1.
guardiansearch.sh can mirror the full public, API, and MCP surface without splitting the stack.
Research, buying flows, link discovery, finance lookups, and other tasks that need live sources.
Use the MCP endpoint for tool-based agents or the JSON API for direct application logic. Guardian also publishes OpenAPI, llms.txt, OpenSearch metadata, and an agent manifest.
Point your client at the MCP endpoint and start calling search tools right away.
Use Guardian Search MCP for retrieval, then open the browser only for pages that truly need interaction.
The Terraform stack can provision guardiansearch.sh as a full mirror of guardianbrowser.sh, including api and mcp subdomains.
{
"mcpServers": {
"guardian_search": {
"transport": "streamable_http",
"url": "https://mcp.guardianbrowser.sh/v1"
}
}
}
POST https://api.guardianbrowser.sh/v1/agent/search
{
"query": "best paper towels",
"vertical": "shopping",
"count": 10,
"engine": "all",
"tool": "search_shopping"
}
curl -sS 'https://api.guardianbrowser.sh/v1/agent/search?q=security%20browser&vertical=all&tool=discover_sites' \
-H 'accept: application/json'
Run the public site, API, and MCP endpoint behind one stack. Keep guardianbrowser.sh as the primary brand today and keep guardiansearch.sh ready as a 1:1 mirror if you ever need to switch names.
The same deployment serves the browser site, the JSON API, and the dedicated MCP hostname.
Provision apex, www, api, and mcp records for the mirror domain and include them in ACM in one plan.
mcp.* serves Streamable HTTP at /v1. The legacy /mcp path redirects to /v1.
Guardian keeps browser automation for browser work and keeps search retrieval in a compact, inspectable interface.
Give them a search interface built for software. Use the browser only when a task truly needs a browser.