orchata_skill

Overview

This skill provides programmatic access to Orchata, a retrieval-augmented generation (RAG) knowledge management platform with tree-based document indexing. Use it to search, browse, and manage Spaces and Documents via MCP tools. It streamlines semantic search, document upload, and hierarchical navigation so you can find precise content quickly.

How this skill works

The skill exposes MCP commands to list and manage Spaces, upload and retrieve Documents, and perform tree-aware semantic queries. Documents are parsed into hierarchical trees (nodes with titles, summaries, page ranges, and content) and indexed for semantic search. Use query_spaces for full content retrieval, smart_query to discover relevant spaces, and tree tools to inspect or read specific sections.

When to use it

• Answer factual questions that require source passages or page ranges
• Discover which knowledge spaces contain relevant information before searching
• Browse large documents by section rather than scanning entire files
• Upload or update corporate documentation and wait for processing to complete
• Audit document processing status (PENDING, PROCESSING, COMPLETED, FAILED)

Best practices

• Default to query_spaces with compact=false for data, numbers, and details — compact=true is for discovery only
• If you don’t know where content lives, run smart_query first, then query_spaces using returned spaceIds
• Only query documents with status COMPLETED; other statuses return no usable content
• When adding content, verify processing by listing documents with status=COMPLETED before relying on results
• Use get_document_tree then get_tree_node to navigate and read specific sections of large documents

Example use cases

• Search across product documentation to extract an API authentication snippet and cite the source and page range
• Discover which legal or policy spaces mention a specific regulation using smart_query, then query those spaces for details
• Upload a batch of onboarding guides to a new Space and confirm all files processed successfully
• Inspect the tree of a long manual to find the installation chapter and read only that node
• Update a document’s content or metadata and append incremental notes while preserving history