MCP Tools Reference

Core Tools (Phase 1)

1. ingest_memory

Ingest data into knowledge graphs from various sources.

# Arguments
{
    "source_path": str,        # Path, URL, or text content
    "source_type": str,        # "file", "url", "text"
    "target_kg": str,          # Target KG name
    "source_name": str,        # Optional human-readable name
    "mode": str = "simple",     # "simple" or "smart"
    "auto_detect_domains": bool = True,
    "auto_create_kgs": bool = True
}

# Usage
result = await client.ingest(
    source_path="https://arxiv.org/abs/2301.07041",
    source_type="url",
    target_kg="KG_Research",
    mode="smart"
)

2. retrieve_memory

Retrieve information from knowledge graphs.

# Arguments
{
    "query": str,              # Search query
    "kg_names": list[str] = None,    # Specific KGs to search
    "top_k": int = 20,         # Number of results
    "mode": str = "hybrid"     # "vector", "structural", "hybrid"
}

3. predict_from_memory

Generate grounded answers with citations.

# Arguments
{
    "query": str,              # Question to answer
    "validate": bool = True,   # Enable answer validation
    "max_context": int = 4000  # Max context tokens
}

Enhanced Tools (Backend Integration)

4. execute_smart_query

Execute enhanced smart query with intelligent routing.

# Arguments
{
    "query": str,                    # Search query
    "mode": str = "auto",            # "auto", "kg_rag", "vector_bypass", "sql"
    "use_vector_bypass": bool = False,
    "enable_reranking": bool = True,
    "max_results": int = 10,
    "kg_filter": list[str] = None,
    "reranker_method": str = "listwise",
    "quality_threshold": float = 0.7,
    "enable_analytics": bool = True,
    "include_explanations": bool = True
}

5. execute_enhanced_sql

Execute enhanced SQL with schema intelligence.

# Arguments
{
    "query": str,                     # Natural language query
    "max_results": int = 10,
    "include_explanation": bool = True,
    "query_type": str = "auto",       # "auto", "select", "aggregate", "join"
    "output_format": str = "summary",   # "rows", "summary", "statistics"
    "include_schema_info": bool = True
}

6. execute_enhanced_reranking

Perform quality-aware document re-ranking.

# Arguments
{
    "query": str,                       # Query for context
    "documents": list[dict],            # Documents to re-rank
    "method": str = "listwise",         # "listwise", "pointwise", "pairwise", "hybrid"
    "quality_filter": str = "auto",     # "auto", "strict", "moderate", "lenient"
    "max_results": int = 10,
    "return_analytics": bool = True,
    "deduplication_threshold": float = 0.85,
    "min_quality_score": float = 0.6
}