MCP Tools

MCP is the preferred agent-facing surface for semantic workflows. An agent asks ShrikeDB for prompts and tool schemas, produces typed requests, and lets ShrikeDB validate and execute them.

Discover and call tools

Send tools/list to the HTTP endpoint /api/v1/mcp to retrieve tool names, descriptions, and argument schemas. Use tools/call with a name and arguments object to invoke a tool. The MCP connection guide includes complete curl requests, authentication, database selection, and response handling.

Tool availability does not imply every operation is enabled or authorized. Check feature settings and the returned diagnostic. In particular, Register, review, and query Parquet distinguishes connection reports, mapping drafts, and source-backed query rows.

Core Semantic Tools

Tool Use

semantic_capabilities

Inspect semantic-query support and active limits.

semantic_catalog

Inspect profiles, datasets, and semantic catalog metadata.

semantic_reason

Resolve a natural-language question into candidate semantic interpretations.

semantic_compile

Compile a typed semantic intent into a validated plan.

semantic_explain

Explain the selected plan, profile evidence, ontology clauses, and query-fabric metadata.

semantic_execute

Execute the typed semantic plan and return rows plus evidence.

Prompt Workflows

Common prompt workflows include:

  • semantic_intent_workflow;

  • domain_pack_workflow;

  • domain_self_config_workflow;

  • entity_categorization_workflow;

  • ontology_grounding_workflow;

  • anomaly_investigation_workflow.

Prompts are XML-structured so the client can produce typed JSON rather than raw query strings.

Onboarding Tools

Tool Use

onboarding_inspect

Inspect a dataset before loading or mapping.

onboarding_load_dataset

Load supported tabular/RDF-like data through the onboarding path.

onboarding_import_ontology

Import ontology data for profile grounding.

onboarding_validate

Validate the proposed dataset/profile/ontology state.

onboarding_status

Check current onboarding state.

Lakehouse Tools

Tool Use

lakehouse_connect_source

Register a database-scoped lakehouse source.

lakehouse_list_connections

List lakehouse connections for the selected database.

lakehouse_connection_status

Inspect source status, freshness, and feature state.

lakehouse_refresh_connection_manifest

Refresh source manifest metadata.

lakehouse_profile_connection

Profile Parquet metadata and sketches.

lakehouse_list_mapping_proposals

List candidate or saved mapping proposals.

lakehouse_get_mapping_draft

Fetch a mapping draft for review.

lakehouse_update_mapping_draft

Edit a mapping draft through governed fields.

lakehouse_validate_mapping_draft

Validate keys, lineage, confidence, and sample oracle checks.

lakehouse_save_mapping

Persist a reviewed draft.

lakehouse_publish_mapping

Activate a reviewed mapping when catalog hash and validation checks pass.

lakehouse_query_execute

Run a lakehouse query or inspection request with provenance.

Safety Notes

MCP semantic tools should not carry raw Cypher/SPARQL text for semantic questions. Direct query modes are separate.

Mutation-capable lakehouse or profile tools require explicit permissions, idempotency checks, expected catalog hashes, and audit evidence.