Three tools with complete input schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Tool names follow verb_noun pattern (fetchIntercept, storage, target) but descriptions are action-list style rather than LLM-optimized narrative. All tools have descriptions (194-250 chars, within baseline range), but many parameters lack descriptions or have minimal guidance. No output schemas documented. Error handling is generic ('Error: {message}') with no recovery guidance. Schemas are properly typed JSON Schema via Zod, but parameter descriptions are sparse or missing entirely.

• Document output schemas for each tool. Example for fetchIntercept: 'Returns {rules: [{id, urlPattern, resourceType, captureBody}], captured: [{requestId, url, method, status, body}], message}'. This enables LLMs to chain tools and extract required fields.
• Add descriptions to all parameters. For storage.domain: 'Cookie domain filter (e.g., ".example.com"). Omit to list all domains.' For storage.expires: 'Cookie expiry as Unix timestamp in seconds (e.g., 1735689600 for 2025-01-01). Omit for session cookie.'
• Replace action-list descriptions with narrative. Example: 'Intercept and inspect network requests at the Chrome DevTools Protocol Fetch domain. Capture requests matching URL patterns, modify responses, and retrieve response bodies. Invisible to page JavaScript.' This tells LLMs what the tool does and when to use it.
• Implement structured error responses. Instead of 'Error: {message}', return {error: 'INVALID_ACTION', message: 'Unknown action: xyz', validActions: ['enable', 'disable', ...], retryable: false}. This guides LLM recovery.
• Add parameter constraints to schema descriptions. For limit: 'Max results to return (1-100, default 50)'. For resourceType: 'Filter by resource type. Valid: Document, Script, XHR, Fetch, Stylesheet, Image, Media, Font, Manifest, Other, SignedExchange, Ping, CSPViolationReport, Preflight, SourceMap, WebSocket, EventSource.'.