Scoring was not performed

• Add explicit output schemas to all tools. Document the return type (object, array, string), all fields, field types, and field descriptions. Example: chroma_query_documents should return {results: [{id, document, distance, metadata}], total_count: number}.
• Expand parameter descriptions to 50-150 characters with actionable constraints. For 'collection_name', state: 'Name of the collection (1-255 alphanumeric chars, underscores allowed, case-insensitive). Used for all subsequent operations on this collection.'
• Add enum constraints to parameters with known valid values. chroma_query_documents 'include' should declare: 'enum': ['documents', 'distances', 'metadata', 'embeddings'] with description of what each includes.
• Add numeric bounds to chroma_query_documents 'n_results': add 'minimum': 1, 'maximum': 1000 to the schema. Document: 'Number of results to return (1-1000, default 10). Larger limits may timeout.'
• Document the structure of chroma_add_documents 'documents' parameter. Specify: 'Array of document objects, each with: {id: string, text: string, metadata?: object}. Maximum 1000 documents per call.'
• Implement confirmation flow for chroma_delete_collection. Add a 'confirm_delete' boolean parameter (default false). When false, return a warning; when true, execute deletion. Description: 'Set to true only after confirming the collection name is correct. This operation is irreversible.'
• Add error recovery guidance to all tool descriptions. Example for chroma_create_collection: 'If the collection already exists, the error will state 'Collection exists'. Use chroma_query_documents to verify, then choose to update or delete first.'