This server exposes only 1 tool (read_graph) with a completely empty input schema ({}). The tool has no visible description in the schema definition, no parameter documentation, and no output schema documentation. The tool name itself is generic and provides no context about what 'graph' is being read or what the operation returns. The codebase shows a client implementation (mcp_client.py) that invokes the tool but does not reveal the server-side tool definition, making proper evaluation impossible. Critical pattern violations: tool-description (missing), schema (completely empty), parameter-descriptions (none), response-shaper (no documented output structure).

• Add a comprehensive tool description (100-200 chars) that explains: What graph is being read? Is it a knowledge graph, task graph, memory graph? When should this tool be called? What is returned?
• Define input parameters with types, descriptions, and constraints. Even if no parameters are required, explicitly document this with 'This tool accepts no parameters.'
• Document the output schema. Include field names, types, and descriptions for every field in the response. If the tool returns a graph structure, describe the node and edge formats.
• Rename the tool to be more descriptive. Consider: 'read_memory_graph', 'fetch_knowledge_graph', 'get_graph_data', or similar, something that tells the LLM what domain the tool operates in.
• Add error handling guidance. If the graph is empty, if a specific graph ID is required but not found, or if the operation times out, return an actionable error message with recovery steps.
• Include the server implementation source (mcp_server_memory.py or equivalent) so the actual tool definition, registration, and schema can be reviewed.

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API