This server has 14 tools but suffers from significant quality issues. While basic schemas are present for most tools, descriptions are sparse or generic, parameter documentation is incomplete, and output schemas are not documented. Tools 1-5 (milvus_tools.py) have minimal descriptions (15-30 chars) that fall below the 50-200 char optimal range. Tools 6-11 have slightly better descriptions but still lack detail about when/why to use them. Parameter descriptions are present but often trivial (e.g., 'Name of the collection' for collection_name). No tool documents its output schema or return structure. Tool #14 is defined in TypeScript (milvus-server.ts) but appears abandoned or duplicate. The server does not validate input constraints (e.g., vector dimensions, collection name formats) and provides no error recovery guidance. Naming is mostly sound (verb_noun pattern) but some tools have overlapping responsibilities (e.g., create_collection vs create_collection_with_doc_id; upload_document_ascii vs upload_document_openai).

• Write full descriptions for each tool (50-200 chars). Example: 'insert_data' should be 'Insert one or more vectors into a Milvus collection. Each vector must match the collection dimension. Returns the number of vectors inserted and any errors per batch.'
• Document the complete output schema for every tool. Example: search should return [{id: int, score: float}, ...], list_collections should return {collections: string[], total: int}
• Consolidate overlapping tools: Merge create_collection and create_collection_with_doc_id into a single tool with optional 'metadata_fields' parameter. Merge upload_document_ascii and upload_document_openai into a single tool with an 'embedding_provider' enum parameter (ascii|openai).
• Add format constraints to vector parameters: 'vectors: array of floats, each float between -1.0 and 1.0, array length <= 10000, each vector dimension must equal collection.dim'
• Add error recovery guidance to descriptions. Example: 'If you get CollectionNotFound, call list_collections() to see available collections. If connection fails, check Milvus URI and retry.'
• Move hardcoded connection details (host='standalone', port='19530') to environment variables or server config. Use pymilvus.connections.configure() with injectable URI.
• Add pagination to list_collections: include 'limit' (default 100) and 'offset' (default 0) parameters. Return {collections: [...], total: int, has_more: bool}
• Remove TypeScript code (milvus-server.ts, milvus-mcp-attempts/) or clarify if it's an alternative implementation. Current setup is confusing.

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

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