AI-native federated database cluster — CLI + MCP server. Specialized truth stores behaving as one governed substrate (typed errors, mutation receipts, policy-enforced surfaces).
16 tools with complete schemas and descriptions. Naming follows verb_noun pattern consistently (cluster_propose_mutation, cluster_validate_mutation, etc.). Descriptions are present and substantive (avg ~120 chars), explaining WHAT each tool does and its state transitions. However, descriptions lack WHEN/WHY guidance and dependency hints. Parameter descriptions are adequate but generic. Output schemas are not documented, LLMs cannot predict return structure. Error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint). Pagination parameters present (limit/offset) but no total_count or next_cursor documented. Risk classification (READ_ONLY vs WRITE) is present but not surfaced to LLM.
Approve a validated mutation command. Transitions command from 'validated' to 'approved' status. Requires approval authority.
Commit an approved mutation command. Transitions command from 'approved' to 'committed' status and applies the mutation to the cluster stores. Returns the receipt.
Discover entities and artifacts in the cluster by text search. Returns matching records with metadata.
Explain why a record was retrieved in a search result. Returns the retrieval explanation with ranking and matching details.
Get the overall status and health of the cluster. Returns cluster metadata, store counts, and health indicators.
List all commands in the cluster. Returns commands with optional filtering by status.
Output schemas not documented. LLMs cannot predict return structure (fields, types, nesting). Forces agents to guess what data is available after each call.
Descriptions lack WHEN/WHY guidance and dependency hints. E.g., cluster_propose_mutation does not explain when to call it vs cluster_validate_mutation, or that validation must follow proposal.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next (e.g., 'If approval denied, try cluster_list_commands to check status').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
List all mutation receipts. Returns receipts with optional pagination.
Propose a mutation command (ingest_artifact, create_entity, update_entity, delete_entity, etc.) to the cluster. Returns a staged command in 'proposed' status.
Read a command by ID. Returns the full command record with status, payload, and lifecycle metadata.
Read derivative index records (search results, metadata). Returns index records with source references.
Read the canonical owner truth for an entity by ID. Returns the full entity record (subject to policy redaction).
Read mutation receipts. Returns receipts for committed mutations with affected IDs and metadata.
Retrieve a bundle of related entities and artifacts for a search query. Returns entities, artifacts, and freshness metadata.
Trace the provenance graph for an entity or artifact. Returns the full provenance graph with nodes and edges showing how the record came to be.
Validate a proposed mutation command. Transitions command from 'proposed' to 'validated' status.
Explain the change history for an entity. Returns a narrative explanation of how the entity came to its current state.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint. Risk classification (READ_ONLY vs WRITE) exists in metadata but is not exposed to LLM via tool annotations.
Pagination incomplete. limit/offset present but no total_count, next_cursor, or has_more documented. LLMs cannot determine if more results exist or how to fetch them efficiently.