A modern multimodal chat workspace for LLM experiments with support for intelligent mode, MCP tool integration, and multi-model comparison.
Chat Studio exposes 3 MCP client tools with critical definition gaps. Tool names lack action verbs (listMcpTools, callMcpTool, invalidateMcpSession use camelCase instead of verb_noun convention). Descriptions are present but minimal (10-40 chars). Input schemas are partially visible: callMcpTool has a basic schema with toolName and toolArguments, but parameter descriptions are generic. No output schemas documented. Error handling is absent, no guidance on recovery or categorization. The server acts as an MCP client wrapper rather than exposing domain-specific tools, limiting practical utility.
Execute an MCP tool with provided arguments
Invalidate the current MCP session
List available MCP tools from configured MCP servers
Tool names violate verb_noun convention. 'listMcpTools' and 'callMcpTool' use camelCase; should be 'list_mcp_tools' and 'call_mcp_tool'. LLMs rely on snake_case action verbs to parse intent.
Descriptions are too short (10-40 chars). 'List available MCP tools from configured MCP servers' lacks context on WHEN to call this vs alternatives, and what structure is returned. Minimum 50 chars recommended.
No output schemas documented. LLMs cannot plan downstream calls or extract data without knowing what listMcpTools returns (array of tool objects? field names?). invalidateMcpSession has no input schema visible.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | 2026-07-28+ | v2 |
callMcpTool parameter 'toolArguments' is typed as generic 'object' with minimal description. No guidance on valid keys, required vs optional fields, or format constraints. LLMs will hallucinate argument structures.
No error handling or recovery guidance. If callMcpTool fails (invalid tool name, bad arguments, timeout), LLM receives no actionable error message. No categorization of retryable vs fatal errors.