browse_docs

This page answers how an AI reads the OpenX docs without leaving the MCP server: what browse_docs takes, how browsing and searching work, and what happens when the docs are down.

Purpose

browse_docs reads every page in the OpenX part of this site's docs menu, laid out as the menu lays it out, and the workspace's own guides, one per kind it accepts, under guides/. A page added to the OpenX menu reaches browse_docs with no other step. The docs site receives nothing from the caller: no token and no header.

Browse

Start at the top with paths set to ["./"] (or leave paths out). Each entry has a path, a title and a one-line abstract that says when to read it. Pass a listed path to open it: a section lists its children, and a page path reads the page. A long page comes in parts, with a NextCursor to pass as cursor. Several paths in one call are answered in order.

Search

Add query to search inside the given paths. The answer lists the best-matching blocks, each with its page path, title and heading. A block is a page's text under one heading. Search is by keyword.

Inputs

Input	Type	Meaning
paths	list of text	Where to look. Omit it, or pass ["./"], for the top level.
query	text	Words to search for inside the paths. Omit it to browse.
cursor	text	The NextCursor of an earlier answer: the next part of a page, or more hits.

Examples

{ "paths": ["./"] }

{ "paths": ["openx/guides/mcp"], "query": "401" }

The answer is plain text with a next: line that says what to do next, and the same data as structured content.

Errors

• A path that is not in the docs answers the nearest valid paths. It is never fetched.
• BROWSE_DOCS_UNAVAILABLE: the docs site did not answer. The answer names the docs URL, and the workspace guides under guides/ still answer.
• BROWSE_DOCS_CURSOR_INVALID: pass a NextCursor exactly as an earlier answer gave it.