Server has 10 tools with complete input schemas and descriptions. Naming follows verb_noun convention (open_session, close_session, fetch, etc.). Descriptions are present and substantive (avg ~150 chars), exceeding the 10-char minimum. However, output schemas are entirely undocumented, no tool declares what fields it returns, forcing LLMs to guess downstream field names. Parameter descriptions are present but lack actionable constraints (e.g., 'timeout' has no min/max; 'extraction_type' enum is documented but 'wait_selector_state' lacks context on when to use each state). Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. Security concern: 'page_action' parameter accepts arbitrary Python code execution, creating injection risk without sanitization guidance.

• Document output schema for every tool. Example for fetch: 'Returns {status: number, content: string, url: string, headers: object, cookies: array}'. This enables LLMs to chain tools and extract the right fields.
• Add min/max constraints to numeric parameters. E.g., 'timeout (milliseconds, 1000 - 300000, default 30000)', 'max_pages (1 - 20, default 5)', 'wait (0 - 60000 ms)'.
• Sanitize and validate 'page_action' parameter. Either restrict to a safe DSL (e.g., 'click selector', 'fill selector value') or document that arbitrary Python is dangerous and requires explicit user approval.
• Add error handling guidance to each tool description. E.g., 'Returns 400 if session_id not found (user-fixable: check list_sessions). Returns 504 on timeout (retryable: increase wait parameter).'
• Clarify when to use HTTP methods (get/post/put/delete) vs browser-based fetch. E.g., 'Use get for simple HTTP requests without JavaScript. Use fetch for pages requiring browser automation.'