MemoraEU SDK provides 9 tools with complete input schemas and descriptions in French. All tools have type-annotated parameters and clear descriptions (avg 60 chars). However, output schemas are not documented in the source code, responses are parsed via model classes (Memory, Fact, SearchResult) but their structure is not visible in the provided code. Tool naming follows verb_noun convention (remember, recall, forget, list_memories). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. Error handling exists (APIError, CryptoError, ConnectionError) but recovery guidance is minimal. Descriptions are functional but lack context on when to use each tool vs. alternatives (e.g., when to call recall vs. list_memories).

• Document output schemas for all tools. For each response type (Memory, Fact, SearchResult, Category), list all fields, their types, and whether they are required. Example: 'Returns Memory object with fields: id (string), content (string, decrypted if secret provided), category (string|null), tags (array[string]), created_at (ISO 8601 string).'
• Add tool annotations to the MCP tool registration. Mark 'forget' and 'invalidate_fact' with destructiveHint=true, 'recall' and 'list_memories' with readOnlyHint=true, and 'remember' with idempotentHint=true (if repeated calls with same content are safe).
• Enhance descriptions with use-case guidance. E.g., 'recall: Semantic search for memories by meaning. Use when the user asks a question or describes a concept. Faster than list_memories for large memory stores.' vs 'list_memories: Retrieve recent memories in chronological order. Use to browse recent activity or when you need all memories in a category.'
• Improve error messages with recovery hints. Instead of 'Déchiffrement échoué', return 'Decryption failed: secret key may be incorrect. Verify the secret parameter matches the one used when storing the memory.' This guides the agent to retry with corrected input.
• Remove or repurpose mcp_sse_url. If the server needs to expose its URL, do so via server metadata or environment variables, not as a tool. If this is meant for client configuration, document it in the README instead.

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

• Relies on HTTP+SSE-only transport (deprecated) - migrate to Streamable HTTP