Albert Code is a bash-based installer/orchestrator with 18 tools, but exhibits critical definition gaps. Tools 1-4 (phase_a, phase_b, phase_run, phase_update) have descriptions but NO visible input schemas in the source code, they are inferred from function signatures in lib/phases.sh, not explicitly registered. Tools 5-18 have partial schemas (some params documented, others missing types). Descriptions are present but often generic or French-only, making LLM selection difficult. No error handling guidance, no output schemas documented, no tool annotations. The server is STDIO-only, capping protocol readiness at 50 regardless of definition quality.

• Explicitly register all 18 tools in an MCP tool definition block (JSON or equivalent) with full input/output schemas. Do not rely on bash function inference. Include tool names, descriptions (English, 100-200 chars), input schema with type and description for every parameter, and documented output schema.
• Add tool annotations: set destructiveHint=true for all apply_* and phase_* tools (WRITE risk); set readOnlyHint=true for check_opencode and conformity-check. This enables agents to reason about side effects.
• Expand descriptions to 100-200 characters. Include: (1) what the tool does, (2) when to use it vs similar tools, (3) what it returns, (4) any prerequisites. Example: 'apply_write: Writes or overwrites a file with dry-run support. Use when you need to create or replace a file atomically. Returns success/failure and file path. Requires write permission on parent directory.'
• Add error handling guidance to all tools. For destructive tools (apply_*), document: 'On failure, returns error code and message. Retryable if network/permission issue; non-retryable if file locked or path invalid. Suggest user fix permissions or path, then retry.' For phase tools, document what to do if VM setup fails (network, disk space, etc.).
• Document output schemas for all tools. Example for apply_write: 'Returns {success: boolean, file: string, size: number, timestamp: string}'. For phase_a: 'Returns {status: string, vm_id: string, resources: {cpus: number, memory: number, disk: number}, skills_synced: number}'.