Both tools have minimal descriptions (under 50 chars) and lack parameter-level documentation. The 'brightsy' tool description is generic ('Proxy requests to an Brightsy AI agent') with no guidance on when to use it or what it returns. Parameters 'role' and 'content' have descriptions but lack type clarity, 'content' is a union of string|array but no guidance on what array items should contain. No output schema is documented. The 'agent-proxy' tool is defined in a test file (test-cursor.ts), not the main server, suggesting it may not be properly registered. No error handling guidance, no pagination support, no idempotency markers.

• Expand tool descriptions to 100-200 characters. Example: 'Send a message to a Brightsy AI agent and receive a response. Use this to query the agent for information or execute multi-turn conversations. Returns the agent's text response.'
• Document the output schema explicitly. Specify that the response is {content: [{type: 'text', text: string}]} so LLMs know what fields to expect.
• Add parameter-level descriptions for 'role' (e.g., 'The sender role: typically "user", "assistant", or "system"') and 'content' (e.g., 'The message content. Pass a string for text, or an array of content blocks for rich content (text, images, etc.).').
• Add error handling guidance: 'If the agent is unavailable, the tool returns an error with a retry suggestion. If the message format is invalid, the tool returns a validation error with the expected format.'
• Move agent-proxy tool registration from test-cursor.ts into index.ts so it is properly discoverable and part of the main server.