Deep Copilot is a VS Code extension, not an MCP server. The codebase shows 24 tools with basic naming (verb_noun pattern mostly followed) but critical gaps: descriptions are extremely terse (5-30 chars, well below the 50-200 char LLM-optimized baseline), input schemas are minimal (only 1-2 params per tool, no type validation visible), and no output schemas documented. Tools like run_shell and apply_patch expose dangerous operations with no error handling guidance, confirmation patterns, or permission gates. The tool definitions appear to be hardcoded in tool-executor.js with no visible JSON Schema validation. No evidence of pagination, structured output formatting, or recovery guidance for failures.

• Expand all tool descriptions to 50-200 characters. Include WHAT the tool does, WHEN to use it, and key prerequisites. E.g., 'Read file contents from disk. Returns UTF-8 text or binary data. Use before editing to understand current state. Fails if file does not exist or is not readable.'
• Document output schemas for all tools. Specify field names, types, and what data they contain. E.g., list_dir should return [{name, type, size, modified_time}]. This enables LLMs to chain tools correctly.
• Add confirmation/dry-run patterns to destructive tools. E.g., run_shell should accept a 'dry_run' param that shows what would execute without running it. apply_patch should show a preview before applying.
• Add parameter descriptions and constraints to schemas. E.g., grep_search.pattern should document 'POSIX extended regex pattern' and provide examples. run_shell.command should warn about shell injection risks.
• Implement error recovery guidance. When a tool fails, return actionable next steps. E.g., 'File not found: /path/to/file. Did you mean: /path/to/file.txt? Available files: [list]' instead of a bare 404.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry