MCP MyZ Search

Poorman's search mcp.

Usage

npx playwright install chromium
npx mcp-myz-search

Command line options

Usage: program [options]

Options:
  --default-provider <provider>  default search provider (choices: "bing", "duckduckgo", "google", default: "bing")
  --port <number>                port number
  --proxy <proxy>                proxy server
  --timeout <ms>                 navigation timeout in ms
  --headed                       headed mode
  --channel <name>               browser channel (default: "chrome")
  --profile <dir>                persistent user data directory
  --human-wait <ms>              wait ms for human to solve challenge
  --stdio                        use stdio transport instead of HTTP
  --log-file <path>              log file path (default: stderr)
  -h, --help                     display help for command

Server Parameters (startup only)

• --port <number> or MYZ_SEARCH_PORT (fallback PORT), default 3000.
• --proxy <http|socks URL> or MYZ_SEARCH_PROXY applied to Playwright.
• --timeout <ms> or MYZ_SEARCH_TIMEOUT for navigation/default timeouts.
• --profile <dir> or MYZ_SEARCH_PROFILE_DIR to persist session/cookies.
• --human-wait <ms> or MYZ_SEARCH_HUMAN_WAIT to allow human challenge solve in non-headless.

MCP HTTP Endpoint

• Streamable HTTP MCP at POST /mcp using @modelcontextprotocol/sdk.
• Tools:
  • search-links args { q: string; limit?: number<=50; provider?: 'bing'|'duckduckgo'|'google' } returns { results: Array<{link,title,snippet}>, total }.
    • Provider defaults to bing, or you can change it via --default-provider switch.
  • extract-page-text args { url: string; selector?: string } returns { text, isContentPage, url, message }.

Providers

• bing
• duckduckgo
• google

Examples

• Custom port + proxy: npx mcp-myz-search --port 3100 --proxy socks5://127.0.0.1:1080
• Persistent profile + human wait: npx mcp-myz-search --headed --profile ~/.myz-bing --human-wait 20000

Notes

• The server reuses a single Playwright browser/context/page and serializes access to improve performance.
• Challenge pages are detected; when not using API, you can allow human solve via --human-wait in non-headless mode.
• Text extraction prefers main, #content, article, or a provided selector; returns a message when content is not found.