This MCP server has critical definitional gaps that would prevent production use. Of 2 tools, only 1 (notify) has a complete schema with proper types and descriptions. The tool_runner tool has a description but lacks visible input parameter type definitions in the source code provided. The notify tool is well-defined (message string with minLength, type enum), but tool_runner's parameters (language, code, timeout) appear to lack TypeScript type annotations in the visible schema. Additionally, neither tool documents output schemas, which is essential for agent composition. Error handling appears absent from both tool definitions, there are no recovery guides or error classifications. The server provides no guidance on when to use tool_runner vs other approaches, and the 'code execution' nature of tool_runner raises severe security concerns with no visible input sanitization, timeout enforcement details, or permission gating. Schema documentation baselines show 100% of A+ tools have documented return types; this server has 0%.

• Add explicit input schema definitions for tool_runner with JSON Schema types. At minimum: language (enum: ['typescript', 'javascript']), code (string, maxLength: 10000), timeout (number, min: 100, max: 60000, default: 5000).
• Document output schema for both tools. notify should return {success: boolean, notification_id: string, timestamp: string}. tool_runner should return {stdout: string, stderr: string, exit_code: number, duration_ms: number} or {error: string, error_type: string} on failure.
• Expand tool_runner description to include: (1) What languages are supported, (2) What Node.js/runtime version, (3) Available imports/dependencies, (4) Max code size, (5) What happens on timeout (returns partial output? error?), (6) Example valid code snippet.
• Add error handling documentation. For tool_runner, specify: 'On syntax error, returns {error: "SyntaxError: <details>"}. On timeout (code runs > <limit>ms), returns {error: "Execution timeout"}. On permission denied (accessing file system), returns {error: "Access denied: sandbox does not permit file I/O"}'.
• Rename tool_runner to execute_typescript or execute_code_in_sandbox to make intent clear from the name alone.
• Add security documentation: 'This tool executes untrusted code in an isolated Node.js VM. It cannot access the file system, network, or process environment. Code output is captured and returned; stderr is included for debugging.' Include specific limits: max runtime 5s, max output 100KB.
• For notify, add enum or clear description of when each notification type (info, success, warning, error) should be used. Example: 'Use "success" for completed operations, "error" for failures requiring user action, "warning" for non-blocking issues, "info" for status updates.'

Inferred effective spec: 2026-07-28+.

• Current-spec patterns present: tool annotations