Mcp Ais

Qdrant List Collections

Lists all available knowledge base collections. Use only when the user asks what data sources or knowledge bases are available, or to verify if a specific collection exists. Not for ERP data queries.

Qdrant Create Collection

Cria uma nova collection no Qdrant. Use vector_size=384 para embeddings BGE-small.

Qdrant Ingest

Converte texto em embedding (BGE-small-en, 384 dims) e insere na collection. Use esta tool para popular o knowledge base — não é necessário gerar vetores manualmente. O payload pode conter metadados como table, description, fields, etc.

Qdrant Search Text

Search the AIS knowledge base using natural language. Use before writing SQL or answering process questions — alone or combined depending on the request. Collections and when to use each: - knowledge_rules -> Query templates, reusable patterns, and business logic built from real usage. Always check here first — reuse what already exists instead of rebuilding from scratch. - knowledge_schema -> ERP table and field structure across all companies and ERPs

Qdrant Upsert

Insere ou atualiza um vetor diretamente (requer vetor pré-computado). Prefira qdrant_ingest para texto.

Qdrant Search

Searches Qdrant using a pre-computed vector. Use only when you already have a float vector and need to query directly. For natural language searches, always prefer qdrant_search_text instead. Administrative/developer use only.

Qdrant Delete

Remove um ponto do Qdrant pelo seu ID. Use com cuidado — operação irreversível.

Qdrant Collection Info

Returns technical metadata about a Qdrant collection (point count, vector size, status). Use only for administrative or diagnostic purposes — e.g., verifying if a collection was populated correctly after ingestion. Not for ERP data queries.

Search Sl8 Docs

Search official Syteline 8 documentation (PDFs, web content, guides) using natural language. Use when the user wants to understand how an SL8 form, process, or module works — not for finding table names or writing queries (use qdrant_search_text + knowledge_schema for that). Examples: "how does the RMA process work in SL8?", "how to configure item lead times?", "what does the costing method affect?". Applies to any company running Syteline 8.

Search Sl10 Docs

Search official Syteline 10 / Infor CSI documentation (PDFs, web content, guides) using natural language. Use when the user wants to understand how an SL10 form, process, or module works — not for finding table names or writing queries (use qdrant_search_text + knowledge_schema for that). Examples: "how does MRP planning work in SL10?", "what is allocated vs on-hand quantity?", "how to process a customer return?". Applies to any company running Syteline 10.

Sql Query

Use this tool to retrieve ERP data for any connected company whenever the user asks about items, parts, orders, inventory, BOMs, jobs, purchasing, shipping, receiving, costing, quality, or any business data. Always follow this flow before writing SQL: 1. Search 'knowledge_rules' first — it contains validated query templates and learned patterns that can be reused directly or adapted. 2. Search 'knowledge_schema' to confirm correct table and field names. 3. If the request also involves unders

Sql Batch

Use this tool for complex multi-step ERP queries requiring CTEs (WITH), DECLARE, temp tables, recursive logic, or multiple dependent steps. Examples: recursive BOM explosion, multi-site inventory comparison, overdue order analysis, serial number tracing, cross-company reporting. Always follow this flow before writing SQL: 1. Search 'knowledge_rules' first — reuse validated templates when available. 2. Search 'knowledge_schema' to confirm correct table and field names. 3. Combine with sl8_docs o