Three tools with complete input schemas and descriptions. ask-qoder has excellent parameter coverage (11 params, all typed and described) with clear permission/sandbox semantics. Descriptions are substantive (100-200 chars, matching baseline). However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling guidance is absent. list-models and list-sessions lack parameter descriptions (empty objects). No tool annotations (readOnlyHint/destructiveHint). Overall solid naming and schema completeness, but missing output documentation and error recovery patterns.

• Document output schema for ask-qoder: {session_id: string, result: object, ok: boolean, code: number, stdout: string, stderr: string, timedOut: boolean, truncated: boolean}. Include field descriptions so LLMs know what to extract for multi-turn resume_session_id.
• Add error recovery guidance to ask-qoder description: 'On timeout, increase timeout_ms or simplify the task. On truncation, reduce output or split into smaller tasks. On qodercli error, verify qodercli is installed and QODERCLI_PATH is correct.'
• Add tool annotations via MCP SDK: ask-qoder should have destructiveHint=true (can write files, run shell); list-models and list-sessions should have readOnlyHint=true.
• Clarify permission parameter precedence in ask-qoder description: 'permission_mode overrides approval_policy and sandbox. If none are set, defaults to QODERCLI_DEFAULT_PERMISSION_MODE (default: dont_ask, read-only).'
• Document list-models and list-sessions output schemas: list-models returns {models: [{name: string, description?: string}]}; list-sessions returns {sessions: [{id: string, created_at: string, status: string}]}.