The rust-analyzer MCP server defines 9 tools with reasonable schemas and descriptions, but falls short of production-grade quality in several dimensions. Most tools have basic descriptions (50-150 chars) and complete input schemas with proper typing, placing them solidly in the fair-to-good range. However, several tools lack depth in their descriptions, parameter documentation does not explain constraints or expected formats, and there is no visible output schema documentation, critical for LLMs planning downstream tool calls. Error handling patterns are absent from the visible code. The server includes both read-only and write operations, but lacks confirmation/dry-run patterns for destructive operations (format, set_workspace). Parameter descriptions are minimal and do not follow the 'actionable constraints' pattern (e.g., 'line must be 0-based, character position 0-based' is stated in the description but lacks examples or guidance on typical ranges).

• Document output schemas for all 9 tools. Show example return objects with field names, types, and what each field represents. E.g., hover should show {type: string, range: {start: {line, character}, end: {line, character}}, contents: string}.
• Expand parameter descriptions to include constraints and examples. E.g., 'Line number (0-based, 0 - <max_lines_in_file>). If the file has 50 lines, line must be 0-49. Character position (0-based, 0 - <max_chars_in_line>).'
• Add a 'When to use' section to tool descriptions. E.g., hover: 'Call this to get the type and documentation of a symbol under the cursor.' definition: 'Call this to navigate to where a symbol is defined.' code_actions: 'Call this to discover refactoring or fix suggestions at a range.'
• Implement confirmation/dry-run for set_workspace. Either require an explicit 'confirm=true' param, or add a 'preview_workspace' tool that shows what would change before committing.
• For format tool, add a 'dry_run' or 'preview' parameter so agents can see the formatted output before applying it destructively.
• Add error handling guidance to each tool. E.g., completion tool: 'Returns an empty list if the position is invalid or the file is not recognized as Rust code. Ensure set_workspace() has been called first and file_path is within the workspace.'
• For symbols tool, add a limit parameter (e.g., max_results: 50) and document pagination or truncation behavior to prevent large responses from exhausting context.
• Clarify file_path resolution. Should it be absolute or relative to workspace? Does it need to exist, or can completion work on unsaved buffers? Document assumptions so agents don't pass invalid paths.