Server has 19 well-named tools with consistent verb_noun patterns (start_session, stop_session, hover, go_to_definition, etc.). All tools have descriptions and input schemas with typed parameters. However, parameter descriptions are minimal (often just the parameter name restated), output schemas are not documented, and error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Composition is strong, tools chain well via session_id, but descriptions lack the depth needed for LLM selection confidence.

• Add detailed parameter descriptions to all tools. For each parameter, explain what it is, its format, valid range/values, and how to obtain it. E.g., 'session_id: Unique identifier returned by start_session(). Use list_sessions() to find active sessions.'
• Document output schemas for all tools. Include the structure of returned objects, field types, and what each field represents. This enables LLMs to chain tools and extract required data.
• Add tool annotations: mark read-only tools with readOnlyHint=true, destructive tools (stop_session, apply_workspace_edit) with destructiveHint=true, and idempotent tools with idempotentHint=true.
• Add error handling guidance to tool descriptions. Document common failure modes (e.g., 'session not found', 'invalid file path') and recovery steps (e.g., 'Call list_sessions() to verify the session exists').
• Restrict or remove raw_lsp tool. If it must exist, whitelist safe LSP methods and validate parameters server-side. Consider replacing with domain-specific tools for common LSP operations.