Two tools with clear verb-noun naming (get_context_pack, get_file_summary) and reasonable descriptions (194 and 47 chars respectively). Input schemas are present and use Zod with type definitions. However, parameter descriptions are sparse or missing context about constraints, ranges, and error conditions. Output schema is not documented, LLMs cannot predict response structure. Error handling is minimal (only get_file_summary returns isError flag). No tool annotations (readOnlyHint, idempotentHint). Composition is sound (single responsibility), but descriptions lack the depth needed for reliable LLM selection and parameter reasoning.

• Document output schema for both tools. For get_context_pack, specify: { files: [{path, tokens, relevance_score}], dependency_graph: {...}, total_tokens, mode }. For get_file_summary, specify: { path, summary: {imports, exports, size_bytes, token_estimate} }.
• Add constraint details to get_context_pack parameters: budget_tokens (range 100 - 100000, default 8000), mode (json for structured, text for markdown, cursor for Cursor IDE format). Explain task format: 'Natural language description of the coding task (10 - 500 chars). More specific tasks yield better results.'
• Expand get_file_summary description to 100+ chars: 'Retrieve a minimal structured summary of a single file, including imports, exports, and token estimate. Use this to preview a file before including it in a context pack. Returns error if file not found.'
• Add readOnlyHint: true to both tool definitions (via server.registerTool options or metadata) to signal to agents that these calls are safe to retry and have no side effects.
• Enhance error responses with recovery guidance. When file not found, return: { error: 'File not found: <path>', suggestion: 'Call get_context_pack with your task to discover relevant files.' }