chat.md exposes a single MCP system tool (system.fetch_mcp_resource) that bridges to other MCP servers' resources. The tool itself has a schema and description, but exhibits significant definition quality gaps: the description is minimal (103 chars, below the baseline average of 194 chars), lacks context on WHEN to use it or WHY, and does not guide recovery if a resource fetch fails. The schema is present and properly typed, but parameter descriptions are terse (serverId: 'Configured MCP server ID that advertised the resource', 61 chars, below baseline param avg of 72 chars; uri: 'Exact MCP resource URI to read from that server', 46 chars). The tool accepts only two required string parameters with no validation constraints (no pattern, no enum, no length limits), making it vulnerable to hallucinated server IDs or malformed URIs. Error handling is absent, no guidance on what to do if a fetch fails (server not found, resource not advertised, network error, etc.). The tool's purpose is singular (fetch a resource) and well-named, but lacks the depth needed to guide LLM selection and recovery.

• Expand tool description to 150 - 250 chars. Example: 'Fetch the full contents of an MCP resource advertised by another configured server. Use this to retrieve documentation, templates, or data files referenced in chat. Returns the resource content and metadata. If the server or resource is not found, try list_mcp_resources() first to discover available options.'
• Add parameter constraints and examples. serverId description: 'The ID of a configured MCP server (e.g. "github", "slack"). List available servers via system diagnostics or configuration.'
• Add URI format guidance: 'The exact resource URI as advertised by the server (e.g. "file:///docs/api.md"). Use list_mcp_resources to find available URIs.'
• Document the response structure: return type, content encoding (text/binary), metadata fields, and any size limits.
• Add error handling: 'If server not found, verify the server ID in MCP configuration. If resource not advertised, call list_mcp_resources() to discover available resources. If fetch times out, the resource may be large, retry or request a smaller subset.'
• Consider adding an optional 'format' parameter (enum: auto, text, json, binary) to guide parsing if the resource is ambiguous.