Two tools present with minimal schema documentation. mcpToolCall has a detailed input schema with 7 parameters (name, version, transportType, toolName, toolArguments, toolCallId, serverUrl, serverId), all with descriptions and type constraints (enum for transportType). gptFunction has 4 parameters (name, description, schema, strict) with basic descriptions. However, neither tool has documented output schemas, critical for LLM planning. Descriptions are present but generic (mcpToolCall: 'Connects to an MCP server and gets a tool call response'; gptFunction: 'Defines a tool, which is a method that the LLM can call'). No error handling guidance, no recovery patterns, no examples of expected outputs. Tool names lack action verbs (mcpToolCall, gptFunction are noun-heavy). No evidence of idempotency, pagination, or composition patterns.

• Rename tools to verb_noun format: 'call_mcp_tool' and 'define_gpt_function'. This signals intent and helps LLMs disambiguate.
• Document output schemas for both tools. Specify what fields are returned (e.g., mcpToolCall returns {result, error, toolCallId}; gptFunction returns {toolId, registered}).
• Expand descriptions to 100-200 characters. Include WHEN to use each tool, what it modifies, and what to do on failure. E.g., 'Calls a remote MCP tool and returns the result. Use this to invoke tools from other MCP servers. Returns {result, error}. On error, check error.code to determine if retryable.'
• Add schema constraints to 'toolArguments' parameter. Document expected structure or provide an enum of valid argument patterns.
• Add error handling guidance. Specify what errors are possible (network timeout, invalid tool name, schema mismatch) and what the LLM should do (retry, ask user, fall back).