Sovereign Brain has 7 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-80 chars, mostly adequate) and input schemas with type info. However, parameter descriptions are minimal or absent, output schemas are undocumented, and error handling is not visible in the code. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Naming is verb-based but generic (read, write, delete, search). The server lacks pagination guidance, field-level documentation, and recovery hints. Average tool score: 52/100.

• Add output schema documentation to all tools. Example for read(): 'Returns: object with fields {value: any, type: string (json|raw)}' or null if key not found.
• Implement tool annotations in MCP registration: mark read/read_full/search/vec_search/graph_neighbors with readOnlyHint=true; mark write with no hint (default); mark delete with destructiveHint=true.
• Expand parameter descriptions with constraints and examples. For write(): 'value: dict, list, or string to store. Dicts/lists are JSON-serialized; strings stored as-is.' For search/vec_search: 'limit/k: max results to return (1-100, default 5). Results truncated if more available.'
• Add error recovery guidance. Document: 'If key not found, returns null, use search() to discover available keys.' 'If vec_search fails (Ollama unreachable), fall back to search() for keyword matching.'
• Document pagination and result limits. Add to search/vec_search descriptions: 'Returns up to limit/k results. For large datasets, use limit=5-20 to avoid context window exhaustion. No cursor/offset pagination, use keyword refinement for narrower results.'