Tool domains

English | 简体中文

Version 1.0 has one built-in read-only catalog: eight top-level domains and fifty-five exact children. This page explains product intent. The generated tool registry is authoritative for formal feature IDs, handler bindings, authorization identifiers, variants, and migration inputs.

Discovery and execution shape

Discover a domain:

{}

Execute one discovered child:

{
  "child_tool": "<exact child name>",
  "arguments": {},
  "manifest_version": "<returned by discovery>"
}

Availability is runtime-specific. A child listed below is part of the public contract, but it is callable only when its version, feature, provider, permission, route, and configuration predicates pass.

doris_catalog — metadata navigation

Child	Purpose	Important behavior
list_catalogs	List internal and external catalogs.	Bounded, ordered, filterable; external visibility follows Doris privileges.
list_databases	List databases in one catalog.	Requires an exact catalog when the route uses multiple catalogs.
list_tables	List tables/views in one database.	Stable filtering and signed cursor pagination.
get_table_context	Assemble table basic, schema, comments, and indexes sections.	schema is always required; optional sections can report partial/unavailable independently.
get_table_size	Return table and bounded partition size evidence.	Uses live statistics/metadata; reports truncation rather than unbounded scans.

get_table_context replaces five pre-1.0 top-level calls. It keeps section provenance and warnings instead of flattening partially available metadata into one ambiguous object.

doris_query — read-only query and diagnostics

Child	Purpose	Important behavior
execute_query	Execute one read-only SQL statement.	SQL guard, bound parameters, timeout/row/byte ceilings, masking, deterministic errors.
explain_query	Return a bounded Doris query plan.	The target SQL must itself pass the read-only guard.
get_query_profile	Retrieve a bounded FE query profile.	Requires FE HTTP/profile evidence and never returns credentials.
diagnose_query_performance	Produce deterministic findings from plan/profile/query evidence.	Evidence-based rules; no invented confidence score.
list_slow_queries	List bounded slow-query evidence.	Requires readable audit history; raw sensitive SQL fields are sanitized.
get_adbc_connection_info	Report Arrow Flight SQL/ADBC readiness.	Advanced, default-off, and valid only after explicit user intent selects ADBC.
execute_adbc_query	Execute a bounded read-only query through ADBC.	Requires explicit_adbc=true; never substitutes for ordinary execute_query; fail-closed on token-bound routes.

Ordinary SQL always uses execute_query. The two ADBC children are an advanced path for requests that explicitly name ADBC or Arrow Flight SQL; their schemas require explicit_adbc=true, and the runtime independently enforces the same condition. The query domain accepts SQL because SQL is the native Doris query interface, but it does not expose DDL, DML, administrative SQL, arbitrary stacked statements, or an unbounded result channel.

doris_cluster — runtime and resource state

Child	Purpose	Important behavior
get_cluster_overview	Summarize deployment, versions, and node health.	Preserves mixed/dead component evidence without treating dead nodes as active gates.
list_cluster_nodes	List FE/BE and relevant component inventory.	Sanitized host/state/version evidence.
list_active_tasks	Inspect active query/load/task progress.	Selects unified or legacy views from live evidence.
get_monitoring_metrics	Retrieve allowlisted bounded metrics.	Metric names, node count, and values are bounded.
get_memory_stats	Inspect memory trackers and use.	Reports source and partial availability.
get_cache_status	Inspect file/cache status.	Selects advanced or base evidence by patch and live metadata.
get_compaction_status	Inspect compaction state.	Uses system table or allowlisted HTTP evidence, with explicit fallback.
get_workload_group_status	Inspect workload-group definitions and metrics.	Requires readable workload-group metadata.
get_compute_group_status	Inspect compute groups.	Availability depends on deployment and readable metadata.
analyze_resource_growth	Analyze bounded recorded resource history.	Requires historical metrics/audit evidence; no speculative forecast.
get_runtime_capabilities	Return a sanitized capability snapshot.	Exposes bounded reasons/evidence, not credentials or private route internals.

doris_pipeline — ingestion and freshness

Child	Purpose	Important behavior
get_ingestion_status	Inspect batch, stream, routine, and insert jobs.	Version-aware job views; bounded collections.
diagnose_ingestion	Diagnose observed ingestion failures/state.	Deterministic rules over returned evidence.
get_materialized_view_status	Inspect MV definitions, jobs, tasks, and compute context.	Selects available metadata paths and reports partial state.
monitor_data_freshness	Report bounded table freshness evidence.	Does not claim freshness without an observable timestamp/source.
analyze_data_dependencies	Infer bounded upstream/downstream dependencies.	Audit-derived runtime evidence; not a complete static lineage proof.

doris_search — text, vector, and hybrid retrieval

Child	Purpose	Important behavior
search_data	Run structured text, vector, or hybrid retrieval.	Validates target indexes/fields, vector dimensions, filters, top-k, and output fields.
preview_text_analysis	Preview Doris tokenization/analyzer output.	Input text and output tokens are bounded.
inspect_search_indexes	Inspect inverted and ANN index metadata.	Does not expose storage secrets or unrelated table properties.
diagnose_search_query	Explain and diagnose search execution.	Requires a validated structured request and bounded plan evidence.

Caller values remain driver-bound. The domain does not accept raw Doris SEARCH DSL fragments or arbitrary filter SQL through the structured search child.

doris_governance — quality, lineage, audit, and access

Child	Purpose	Important behavior
analyze_columns	Analyze bounded column statistics and quality evidence.	Column count and sampling are bounded; Doris permission remains final.
analyze_table_storage	Inspect partitions, indexes, and storage shape.	Sensitive locations/properties are removed from model-facing output.
get_lineage_capability_status	Explain which lineage path is active and why.	Separates native, audit-primary, degraded fallback, and unavailable states.
trace_column_lineage	Trace attributable bounded lineage edges.	Native provider on eligible routes; conservative audit inference otherwise.
analyze_data_access_patterns	Summarize bounded audit access evidence.	Raw SQL/client/auth data is sanitized.
get_recent_audit_logs	Return bounded recent audit events.	Time window/result count are capped; sensitive fields redacted.
list_udfs	Inspect visible UDF metadata.	Version-aware function family metadata; no function execution.
get_auth_mapping_status	Inspect sanitized LDAP/OIDC role mapping readiness.	Does not expose bind credentials, rules containing secrets, or raw provider errors.

Native lineage requires both an eligible Doris version and a healthy queryable companion store. Version 4.0.6+ alone does not make it callable.

doris_lakehouse — external and semi-structured data

Child	Purpose	Important behavior
inspect_external_catalog	Inspect visible external catalog metadata and readiness.	Properties and locations are sanitized.
inspect_lakehouse_table	Inspect table format, snapshots, partitions, lifecycle, and pushdown evidence.	4.1 lifecycle facets are reported separately from base metadata.
inspect_variant_column	Inspect Variant type and bounded shape evidence.	Sample values and sensitive paths are excluded; advanced 4.1 facets are capability-gated.

doris_semantic — optional semantic consumers

Child	Purpose	Important behavior
list_semantic_models	List bounded reviewed models from the configured provider.	Domain is default-off and requires explicit provider enablement.
get_semantic_model_summary	Return a bounded summary for one exact model_ref.	No prompt-based model guessing.
get_semantic_context	Build deterministic read-only grounding context.	Requires exact model, route-aware mapping, and Doris visibility.
get_semantic_mapping_status	Explain private Doris binding readiness.	Reports status/reasons without exposing the private binding manifest.
list_metricflow_models	List configured MetricFlow model references.	Provider is default-off; model selection never uses prompt guessing.
get_metricflow_status	Return provider/model validation status.	Requires one exact model_ref.
list_metricflow_metrics	List bounded metrics and optional dimensions.	Requires one exact model_ref; supports bounded filtering.
get_metricflow_group_bys	Return valid group-by dimensions for selected metrics.	Provider validates the metric set against the exact model.
list_metricflow_saved_queries	List bounded saved-query metadata.	Does not execute saved queries.
get_metricflow_dimension_values	Compile and execute a bounded dimension-value query.	Provider compiles Doris SQL; MCP SQL guard and Query runtime execute it.
compile_metricflow_query	Compile a structured MetricFlow request to Doris SQL.	Compile-only; returns no query result and does not bypass the SQL guard.
execute_metricflow_query	Compile and execute a bounded MetricFlow request.	Execution always returns to DorisQueryRuntime for routing, RBAC, limits, audit, and redaction.

Ossie owns semantic definitions. MetricFlow owns metric semantics and query compilation. The MCP Server is a governed consumer: it does not author either model, does not guess a model, and does not permit a provider to execute Doris SQL outside the MCP Query runtime. See MetricFlow integration.

Reserved and extension surfaces

• doris_admin is intentionally not registered.
• Custom providers are explicit extensions and not counted in 8/55.
• MCP resources and prompts remain separate protocol surfaces; they are not hidden child tools.
• Pre-1.0 tool names are not callable aliases.

See Capability availability for how each child becomes callable=true or callable=false.