Cryo MCP Server

Instructions

Implementation Reference

• cryo_mcp/server.py:658-693 (registration)

  Registration and handler stub for the 'query_sql' MCP tool. Delegates execution to execute_sql_query in sql.py. Input schema defined by function parameters and docstring.
  @mcp.tool() def query_sql( query: str, files: Optional[List[str]] = None, include_schema: bool = True ) -> Dict[str, Any]: """ Run a SQL query against downloaded blockchain data files IMPORTANT WORKFLOW: This function should be used after calling query_dataset to download data. Use the file paths returned by query_dataset as input to this function. Workflow steps: 1. Download data: result = query_dataset('transactions', blocks='1000:1010', output_format='parquet') 2. Get file paths: files = result.get('files', []) 3. Execute SQL using either: - Direct table references: query_sql("SELECT * FROM transactions", files=files) - Or read_parquet(): query_sql("SELECT * FROM read_parquet('/path/to/file.parquet')", files=files) To see the schema of a file, use get_sql_table_schema(file_path) before writing your query. DuckDB supports both approaches: 1. Direct table references (simpler): "SELECT * FROM blocks" 2. read_parquet function (explicit): "SELECT * FROM read_parquet('/path/to/file.parquet')" Args: query: SQL query to execute - can use simple table names or read_parquet() files: List of parquet file paths to query (typically from query_dataset results) include_schema: Whether to include schema information in the result Returns: Query results and metadata """ from cryo_mcp.sql import execute_sql_query return execute_sql_query(query, files, include_schema)
• cryo_mcp/sql.py:99-255 (handler)

  Core handler function that executes the SQL query using DuckDB. Handles file registration as views, query execution, schema extraction, and result formatting.
  def execute_sql_query( query: str, files: Optional[List[str]] = None, include_schema: bool = True ) -> Dict[str, Any]: """ Execute a SQL query against specified parquet files. Args: query: SQL query to execute files: List of parquet file paths to query. If None, will use all files in the data directory. include_schema: Whether to include schema information in the result Returns: Dictionary with query results and metadata """ data_dir = get_data_directory() conn = create_connection() try: # Determine which parquet files to use parquet_files = [] if files: for file_path in files: path = Path(file_path) if path.exists() and path.suffix == '.parquet': parquet_files.append(path) else: print(f"Warning: File not found or not a parquet file: {file_path}") else: # If no files provided, use all parquet files in the data directory parquet_files = list(data_dir.glob("**/*.parquet")) if not parquet_files: return { "success": False, "error": "No parquet files available. Download data first with query_dataset." } # Register temporary views for datasets if needed has_registered_views = False try: # Check if the query might be using direct table references without read_parquet() potential_tables = extract_tables_from_sql(query) # Create views for potential table names that aren't using read_parquet for table_name in potential_tables: if not ("read_parquet" in query.lower() and table_name.lower() in query.lower()): # Match files to table name more precisely # First, look for exact dataset name match (e.g., "blocks" in ethereum__blocks_*.parquet) dataset_pattern = f"__{table_name.lower()}__" exact_matches = [f for f in parquet_files if dataset_pattern in str(f).lower()] # If no exact matches, try looser matching if not exact_matches: # Try matching at word boundaries to avoid partial matches matching_files = [] for f in parquet_files: file_lower = str(f).lower() # Match dataset name patterns like ethereum__blocks_* or *_blocks_* if f"__{table_name.lower()}__" in file_lower or f"_{table_name.lower()}_" in file_lower: matching_files.append(f) # Also match if it's just the table name at the start of the filename elif f"/{table_name.lower()}_" in file_lower or f"/{table_name.lower()}." in file_lower: matching_files.append(f) else: matching_files = exact_matches if matching_files: # Create a combined view from all matching files conn.execute(f"DROP VIEW IF EXISTS {table_name}") if len(matching_files) == 1: # If only one file, create a simple view conn.execute(f"CREATE VIEW {table_name} AS SELECT * FROM '{matching_files[0]}'") print(f"Registered view '{table_name}' for file: {matching_files[0]}") else: # If multiple files, create a UNION ALL view to join all files union_query = " UNION ALL ".join([f"SELECT * FROM '{file}'" for file in matching_files]) conn.execute(f"CREATE VIEW {table_name} AS {union_query}") print(f"Registered view '{table_name}' for {len(matching_files)} files using UNION ALL") has_registered_views = True # Execute the query print(f"Executing SQL query: {query}") result = conn.execute(query).fetchdf() # Convert to records format for easier JSON serialization records = result.to_dict(orient="records") # Get schema information if requested schema_info = None if include_schema and not result.empty: schema_info = { "columns": list(result.columns), "dtypes": {col: str(dtype) for col, dtype in result.dtypes.items()} } # Track how the files were used file_usage = {} if has_registered_views: for table_name in extract_tables_from_sql(query): # Use the same matching logic as above dataset_pattern = f"__{table_name.lower()}__" exact_matches = [f for f in parquet_files if dataset_pattern in str(f).lower()] if not exact_matches: matching_files = [] for f in parquet_files: file_lower = str(f).lower() if f"__{table_name.lower()}__" in file_lower or f"_{table_name.lower()}_" in file_lower: matching_files.append(f) elif f"/{table_name.lower()}_" in file_lower or f"/{table_name.lower()}." in file_lower: matching_files.append(f) else: matching_files = exact_matches if matching_files: file_usage[table_name] = { "files": [str(f) for f in matching_files], "combined": len(matching_files) > 1 } return { "success": True, "result": records, "row_count": len(records), "schema": schema_info, "files_used": [str(f) for f in parquet_files], "used_direct_references": has_registered_views, "table_mappings": file_usage if file_usage else None } except Exception as e: # Handle query-specific errors error_msg = str(e) print(f"SQL query error: {error_msg}") return { "success": False, "error": error_msg, "files_available": [str(f) for f in parquet_files] } except Exception as e: # Handle connection and setup errors return { "success": False, "error": str(e) } finally: # Clean up any registered views if has_registered_views: for table_name in extract_tables_from_sql(query): try: conn.execute(f"DROP VIEW IF EXISTS {table_name}") except: pass conn.close()
• cryo_mcp/sql.py:34-74 (helper)

  Helper function to list available parquet tables/files for SQL querying, used by list_available_sql_tables tool.
  def list_available_tables() -> List[Dict[str, Any]]: """List all available tables from downloaded data files.""" data_dir = get_data_directory() # Find all parquet files in the data directory (including the latest subdirectory) parquet_files = list(data_dir.glob("**/*.parquet")) tables = [] for file_path in parquet_files: # Extract dataset name from filename name = file_path.stem.split("__")[0] if "__" in file_path.stem: name = file_path.stem.split("__")[0] else: # Try to extract from other naming patterns name_match = re.match(r'([a-z_]+)_', file_path.stem) if name_match: name = name_match.group(1) else: name = file_path.stem # Get file stats stats = file_path.stat() # Try to extract block range from filename block_range = "" blocks_match = re.search(r'blocks__(\d+)_to_(\d+)', str(file_path)) if blocks_match: block_range = f"{blocks_match.group(1)}:{blocks_match.group(2)}" tables.append({ "name": name, "path": str(file_path), "size_bytes": stats.st_size, "modified": stats.st_mtime, "block_range": block_range, "is_latest": "latest" in str(file_path) }) return tables
• cryo_mcp/sql.py:17-33 (helper)

  Helper to create DuckDB connection used by SQL execution.
  def create_connection(read_only: bool = False) -> duckdb.DuckDBPyConnection: """Create a DuckDB connection with appropriate settings.""" # In-memory database can't be read-only, so we always use read_only=False conn = duckdb.connect(database=":memory:", read_only=False) # Configure DuckDB settings for performance and safety conn.execute("SET memory_limit='4GB'") conn.execute("SET max_expression_depth=10000") # Note: query_timeout_ms setting might not be available in all DuckDB versions try: conn.execute(f"SET query_timeout_ms={DEFAULT_QUERY_TIMEOUT * 1000}") except Exception: pass # Ignore if setting doesn't exist return conn
• cryo_mcp/sql.py:268-313 (helper)

  Helper function to get schema and sample data for a table/file, used by get_sql_table_schema tool.
  def get_table_schema(file_path: str) -> Dict[str, Any]: """ Get schema information for a parquet file. Args: file_path: Path to the parquet file Returns: Dictionary with schema information """ conn = create_connection() try: path = Path(file_path) if not path.exists() or path.suffix != '.parquet': return { "success": False, "error": f"File not found or not a parquet file: {file_path}" } # Register a temporary view for the file conn.execute(f"CREATE VIEW temp_view AS SELECT * FROM '{file_path}'") # Get schema info schema_result = conn.execute("SELECT column_name, data_type FROM information_schema.columns WHERE table_name='temp_view'").fetchdf() # Get sample data sample_data = conn.execute("SELECT * FROM temp_view LIMIT 5").fetchdf() # Get row count (might be expensive for large files) row_count = conn.execute("SELECT COUNT(*) as count FROM temp_view").fetchone()[0] return { "success": True, "file_path": file_path, "columns": schema_result.to_dict(orient="records"), "sample_data": sample_data.to_dict(orient="records"), "row_count": row_count } except Exception as e: return { "success": False, "error": str(e) } finally: conn.close()