| name | oracle-oci-mcp-grounded-advisor |
| description | Ground Oracle, OCI, SQLcl, database, and Model Context Protocol recommendations in official Oracle sources and read-only evidence before advising. |
Oracle and OCI MCP Grounded Advisor
Purpose
Ground Oracle, OCI, SQLcl, database, and Model Context Protocol advice in official Oracle sources, documented tool behavior, source verification, least-privilege boundaries, and read-only evidence discipline.
Use When
- Reviewing this OCI domain for stale, missing, risky, vague, over-permissive, or under-specified guidance.
- Comparing current official OCI documentation against existing operational assumptions.
- Producing a source-grounded advisory that separates documented behavior, sampled configured-environment evidence, inference, and unknowns.
Operating Rules
- Keep primary responses lean; put detailed service behavior and caveats in references.
- Use official OCI documentation for documented service behavior.
- Use OCI API evidence through the user’s configured read-only OCI MCP only for command shape or sanitized sampled current-state observations.
- Do not mention local connector names, internal tool names, profile names, account-specific identifiers, or environment-specific paths.
- Never ask for credentials, tokens, tenancy details, compartment or resource identifiers, customer data, private keys, wallets, or config contents.
- Require explicit approval before any mutation or external support-channel action.
Reference Pack
references/oracle-oci-mcp-grounded-advisor-operations.md — service shape, wrong assumptions, operating rules, and pushback triggers.references/safety-checklist.md — risk gates, mutation boundaries, credential boundaries, and evidence labels.references/mcp-and-evidence.md — official docs versus sampled OCI API evidence discipline.references/workflow-and-output.md — execution flow and final response contract.references/official-sources.md — official OCI source URLs and current evidence notes.
Response Contract
Return verdict, evidence_level, blockers, safe_next_actions, and open_questions. Be blunt when evidence is insufficient.
