This server has critical definition quality gaps. Tool descriptions are minimal (5-30 chars, well below the 194-char baseline). Parameter descriptions are sparse or missing entirely. No input schemas are visible in the source code, only OpenAI function definitions with bare property names and no type constraints. The 'plan' tool is particularly problematic: it invokes the LLM internally, creating a nested agent loop with no clear output schema. Error handling is minimal (generic try-except blocks returning 'Error: {str(e)}'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. No pagination support on tools that could return large results (glob, grep). The server is not an MCP server at all, it's a standalone Python agent using OpenAI's function calling, not the MCP protocol.

• Migrate to MCP protocol. Implement an MCP server using the Python SDK (mcp package) with proper transport (HTTP+SSE or Streamable HTTP), stateless request handling, and per-request _meta with logLevel.
• Expand tool descriptions to 50-200 chars. Example: 'execute_bash: Run a shell command and return stdout+stderr. Use for system operations, file checks, or build tasks. Timeout: 30s. Returns combined output or error message.' Include when to use, what it returns, and any limits.
• Add parameter descriptions for all inputs. Example: 'path: Absolute or relative file path. Supports glob patterns (e.g. *.txt). Max 4096 chars.' Include format, constraints, and examples.
• Add JSON Schema type constraints to all parameters. Use minLength, maxLength, enum, pattern, and minimum/maximum. Example: 'command: {type: string, minLength: 1, maxLength: 2048, description: "Shell command to execute. Avoid pipes; use separate tools instead."}'
• Remove or redesign the 'plan' tool. If task decomposition is needed, implement it as server-side logic (not exposed as a tool) or use a separate planning endpoint. Do not nest LLM calls inside tools, this breaks observability and creates infinite loops.