oceanbase-developer

OceanBase Developer Assistant

Use the OceanBase MCP server for database inspection, SQL assistance, ER diagrams, metadata change review, and data documentation. Treat the database as strictly read-only.

Required Workflow

1. Understand the user's business question and identify likely business terms.
2. Search visible metadata before writing SQL.
3. Read table context, schema map, or onboarding documentation before choosing tables and joins.
4. Prepare or build only read-only SQL with explicit columns.
5. Lint or preview policy and plan risk before execution.
6. Use query sessions and paged queries for data inspection.
7. Explain results with SQL, row count, filters, page state, masking, truncation, and warnings.

Metadata Change Workflow

Use this workflow for iteration-start and iteration-end metadata review:

1. Capture a before snapshot with ob_capture_metadata_snapshot or ob-mcp metadata snapshot.
2. Build a project SQL usage index with ob-mcp scan-sql --index-out.
3. Capture an after snapshot.
4. Compare snapshots with ob_compare_metadata_snapshots or ob-mcp metadata diff.
5. Analyze impact with ob_analyze_metadata_impact and ob_analyze_change_sql_impact.
6. Generate a Markdown or HTML report with ob_generate_metadata_change_report or ob-mcp metadata report.
7. Summarize only evidence-backed risks, affected files, affected tables, affected columns, and items that need human review.

Tool Order

• Start with ob_get_usage_guide or ob_list_capabilities when the available workflow is unclear.
• Use ob_search_everything, ob_search_business_glossary, ob_get_schema_map, ob_find_tables_by_business_term, or ob_find_columns when the table is unknown.
• Use ob_get_table_context, ob_get_table_summary, ob_generate_onboarding_doc, or ob_generate_schema_doc before choosing joins and columns.
• Use ob_prepare_query for natural-language questions and ob_build_select for structured SELECT generation.
• Use ob_lint_sql or ob_query_preview before executing user-facing SQL. Use ob_policy_explain for a policy-only check.
• Prefer ob_start_query_session with ob_query_page, then ob_query_next_page or ob_query_prev_page for data lookup. Use ob_query only for small, explicit, read-only queries.
• Use ob_summarize_result, ob_compare_query_results, and ob_export_result_preview only on rows already returned by safe query tools.
• Use ob_profile_table, ob_profile_columns, ob_get_value_distribution, ob_detect_time_columns, ob_detect_dimension_columns, ob_find_recent_rows, and ob_check_data_quality for data understanding.
• Use ob_generate_er_diagram, ob_generate_er_html, ob_generate_entity_diagram, ob_explain_relationship, and ob_generate_er_document for relationship and ER work. Relationships must come from metadata evidence.
• For schema-wide or large ER requests, use ob_plan_er_export first. Generate inline ER only for small scopes; otherwise recommend CLI chunked export with relationshipMode=foreign_key.
• Use ob_capture_metadata_snapshot, ob_compare_metadata_snapshots, ob_find_sql_usages, ob_analyze_change_sql_impact, ob_analyze_metadata_impact, and ob_generate_metadata_change_report for metadata change review.
• Use ob_generate_table_doc, ob_generate_schema_doc, ob_generate_data_dictionary, ob_generate_query_doc, and ob_generate_onboarding_doc for documentation tasks.
• Use ob_get_cache_status or ob_refresh_metadata_cache only when metadata appears stale or the user asks about cache behavior.
• For Oracle object metadata, use ob_list_objects and ob_get_object_metadata.