Server has 3 tools with complete Zod schemas and descriptions, but descriptions are verbose (200-400+ chars) and lack LLM-optimized structure. Tool names lack clear verb-noun patterns (import_plugin, dev_plugin are ambiguous about what 'import' and 'dev' mean in context). Parameters are described but descriptions are dense Chinese text without actionable constraints or format guidance. No output schemas documented. Error handling is absent, no guidance on recovery, retryability, or what to do on failure. No tool annotations (readOnlyHint, destructiveHint). Composition is weak: import_plugin and dev_plugin overlap significantly (both open browser, import plugin, listen to console); should be split into separate concerns.

• Rename tools to clearer verb-noun patterns: 'upload_plugin' (instead of import_plugin), 'debug_plugin' (instead of dev_plugin), 'get_console_logs' (already good). Verify names match what the LLM will infer.
• Document output schemas for all three tools. E.g., import_plugin should return {success: boolean, pluginId: string, message: string}. get_console_logs should return {logs: [{timestamp: string, level: string, message: string}], total: number}.
• Add error handling guidance: 'If plugin upload fails, check that pluginPath points to a valid .eext file. If browser detection fails, pass browserPath explicitly.' Include retryability classification.
• Shorten and restructure descriptions to 50-150 chars with actionable constraints. E.g., 'timeout: Maximum wait time in seconds (1-3600, default 300). Longer timeouts allow more time to detect errors.'
• Split import_plugin and dev_plugin: create separate 'upload_plugin' and 'monitor_console_logs' tools so agents can compose them independently. Add tool annotations (destructiveHint for upload, readOnlyHint for monitor).