Paths Le

What paths-le MCP server does

The paths-le MCP server gives MCP-compatible agents a focused path-extraction operation. Its main tool, extract_paths, examines content supplied in the request and identifies paths written in that content. Results classify each match as a file, relative path, absolute path, or URL, and include its 1-based line and column.

The parser handles common structured and code-oriented formats, including JavaScript, TypeScript, JSON, JSONC, YAML, TOML, CSV, HTML, CSS, and environment files. JavaScript and TypeScript import analysis excludes npm package names and supports imports spanning multiple lines. HTML and CSS handling covers values such as src, href, srcset, url(), and @import. For other text, the engine uses a scan that looks for quoted filenames and text containing path separators.

How it works

The paths-le MCP server receives document content rather than a file location. Optional request fields identify the format and filename, control duplicate handling, and set a maximum number of results. Paths are returned exactly as they appear in the input; the server does not resolve them against a workspace, follow symlinks, access local files, or contact external services.

This makes the MCP interface suitable for agents that already have file contents available through their host environment. It is distinct from the editor and workspace scanning features described for the broader Paths-LE project: the MCP server itself performs content extraction and does not scan a folder tree.

Setup and configuration

Install or run the Node package with npx:

npx -y paths-le-mcp

A standard MCP configuration can start it over stdio:

{
  "mcpServers": {
    "paths-le": {
      "command": "npx",
      "args": ["-y", "paths-le-mcp"]
    }
  }
}

You can also install the package globally and configure the paths-le-mcp executable directly. Pinning a package version is supported when a host should avoid automatically following later releases. No environment variables, API key, or server-specific configuration is required.

Tools and capabilities

The server exposes one documented operation:

• extract_paths(content, format?, filename?, dedupe?, maxResults?) — extracts matching paths and reports their type and source position.

The optional format and filename inputs can help the engine apply format-specific handling. Deduplication and a maximum-result value let callers constrain the returned data. The same extraction engine is also available through the VS Code extension, but using the MCP package does not require that extension.

Limitations and notes

The paths-le MCP server only processes content provided by the caller. It does not read unsaved or saved files, inspect a workspace, resolve relative paths, or verify that a referenced path exists. It also makes no network requests of its own. Returned values therefore describe the spelling and location of matches in the supplied document, not their resolved destination.

For formats without a dedicated parser, results come from text scanning and may reflect path-like text rather than a semantically validated reference. Agents needing a project-wide report must use the separate editor or CLI workflows documented by the project, rather than relying on the MCP operation alone.