PublicWWW

What PublicWWW MCP server does

The PublicWWW MCP server exposes PublicWWW’s website-source search through MCP. An assistant can look for exact strings, combinations of terms, exclusions, country domains, or content on inner pages. Searches can help identify websites that load a particular JavaScript file, use a framework, include a tracking identifier, or contain a widget.

Results can be saved as clusters: named collections of domains that can be reviewed, combined, or processed further. The server also supports extracting information from the source of sites in a cluster, including contact details, social profile links, and selected platform or application identifiers.

How it works

The service is available as a remote Streamable HTTP MCP endpoint:

https://api.publicwww.com/mcp

It returns JSON answers and does not use sessions. Clients can authenticate with OAuth where supported, or send a PublicWWW API token as a bearer token. The token can be stored in PUBLICWWW_TOKEN.

Queries use PublicWWW syntax. Quoted text searches for an exact string, terms separated by spaces must both occur, a pipe expresses alternatives, and a minus excludes a term. Filters such as site:de restrict domains, while depth:all includes inner pages instead of only home pages.

Setup and configuration

Claude web and desktop, ChatGPT, and Claude Code can authenticate through OAuth. Claude Code can also be configured with its MCP command-line setup. Cursor, VS Code, Gemini CLI, and Codex CLI have documented installation paths in the repository. Other MCP clients can use a client-specific configuration from the repository’s clients/ directory.

For token-based access, issue a token from the PublicWWW profile API page and provide it through the client’s authorization-header configuration or the PUBLICWWW_TOKEN environment variable. Cursor reads that variable, VS Code stores the token in secret storage, and Gemini CLI stores it in the system keychain after the initial prompt.

A free account can connect and read the service documentation. The search, account, and cluster operations require a paid plan and are subject to PublicWWW limits.

Tools and capabilities

The PublicWWW MCP server provides ten tools:

• search runs source-code searches with query, row, page, and snippet options.
• account reports the plan, remaining searches, and extraction points for the day.
• read_docs returns PublicWWW documentation and pricing information in Markdown.
• Cluster tools list, view, create, combine, rename, delete, and extract from saved clusters.

Clusters can be created from a search or a supplied domain list. Existing clusters support and, or, and diff combinations. Extraction accepts built-in presets such as email, phone, WhatsApp, Telegram, Facebook, Instagram, LinkedIn, YouTube, GitHub, Google Play, and App Store. Custom extraction uses a regular expression whose first capture group supplies the value.

Documentation pages are also available as MCP resources under publicwww://docs/<page>. An account can retain up to 100 clusters, and extraction consumes the account’s daily extraction points.

Limitations and notes

Operational use depends on the PublicWWW plan and its daily search and extraction allowances. Deleting a cluster is permanent. Extraction runs against the page source of sites in a cluster, so the returned values depend on what those pages contain and on the selected preset or regular expression.

For automated scripts, the repository points to PublicWWW’s REST API as a simpler option and identifies client libraries for Python, JavaScript, PHP, Go, and Ruby. The MCP endpoint is intended for MCP-compatible assistants and tools.