Browse AI

Visit browseai.dev/dashboard and sign up for a free account. Copy your API key — it begins with bai_ and is required for all API and MCP server requests.

The browseai-dev package includes a setup command that automatically configures the MCP server for Claude Desktop, Cursor, and Windsurf.

npx browseai-dev setup

Alternatively, add the server entry directly to your claude_desktop_config.json with your BROWSE_API_KEY set as an environment variable.

{
  "mcpServers": {
    "browse-ai": {
      "command": "npx",
      "args": ["-y", "browseai-dev"],
      "env": {
        "BROWSE_API_KEY": "bai_YOUR_KEY_HERE"
      }
    }
  }
}

Restart Claude Desktop (or your client) to load the Browse AI server. Available tools include: browse_search, browse_answer, browse_extract, browse_open, browse_compare, browse_clarity, and session management tools.

For programmatic use in Python scripts or notebooks, install the browseaidev package.

pip install browseaidev

Use the SDK to run evidence-backed queries at different depth levels: fast (single pass), thorough (up to 3 confidence-gated passes), or deep (multi-step agentic research).

from browseaidev import BrowseAIDev

client = BrowseAIDev(api_key="bai_xxx")
result = client.ask(
    "What are the key differences between CRISPR-Cas9 and base editing?",
    depth="thorough"
)
print(result.answer)
print(f"Confidence: {result.confidence:.0%}")