AWS S3 Tables MCP: Documentation
Updated:
Connection
Part of the awslabs/mcp monorepo — installed as the standalone s3-tables-mcp-server package (stdio), run via uvx or pip install. Auth uses standard AWS credentials (an IAM role, profile, or access keys) with S3 Tables permissions.
Tools
Structure management: list_table_buckets, create_table_bucket, list_namespaces, create_namespace; tables: list_tables, create_table, rename_table, create_table_from_csv; data and metadata: query_table (read-only SQL), append_data, get_table_metadata, get_maintenance_config, get_resource_policy, get_bucket_metadata.
Agent flow
"Create a table from this sales-log CSV", "calculate monthly sales in the sales table", "show me the schema and metadata for the events table", "what namespaces exist in this table bucket?".
Important
The server works only with S3 Tables — a specialized Apache Iceberg-based analytical format, not regular S3 objects (for that, you'd need a different tool from awslabs/mcp or general AWS MCP coverage). Don't confuse it with general-purpose S3 file storage.
Pricing
The MCP server is free, open source (Apache-2.0, part of awslabs/mcp). The S3 Tables service itself is billed by AWS for storage volume, requests, and maintenance (compaction).