v0.1.5

v0.1.5 – Safe Query Execution & Read‑Only Hardening

This release introduces the first iteration of safe query execution plus substantial security hardening, formatting improvements, and CI/test infrastructure upgrades.

✨ New Features

• Query Execution (run_sql_query): Execute parameterized, strictly validated read-only SQL (single SELECT / WITH) returning structured JSON results. You can also execute write operations (DDL/DML), see the security improvements section below.
• Related Tables Discovery: Improved cross‑schema foreign key resolution (falls back to actual owner when initial schema context yields no relations).

🔐 Security & Safety Improvements

• Read‑Only Mode enforcement: Unified guard blocks non-SELECT statements early; write ops only possible when READ_ONLY_MODE="0".
• Robust SQL parsing using sqlparse replaces regex approach (prevents stacked statements, semicolon injection, and comment-based bypasses).
• wrap_untrusted improved (XML escaping + clearer boundaries) to reduce prompt-injection risk.
• Defensive token / permission validation adjustments.

🧪 Testing & CI

• Added GitHub Actions workflows: unit tests on every push/PR; integration tests gated after unit success.
• Expanded integration coverage: injection blocking, formatting, related table resolution.

🛠 Refactors & Internal Changes

• Centralized cursor management and explicit commit control—only commits on explicit DML/DDL when write mode enabled.
• Externalized and enhanced result formatter (auto column width, consistent padding).
• Unified query executability assertion method consolidating permission + type checks.
• Relocated wrap_untrusted into db_context.utils for reuse.

🐛 Fixes

• Accurate cross‑schema foreign key detection in get_related_tables using r_owner plus fallback lookup.
• Improved error handling in run_sql_query and broader exception handling paths.

📦 Dependency & Metadata Updates

• Added sqlparse dependency for safe parsing.
• Version bumped to 0.1.5.

📊 Output & Usability Enhancements

• JSON query output now includes ordered columns, structured rows, and row_count.

⚠️ Notes / Potential Breaking Changes

• Stricter read-only checks may now reject ambiguous multi-statement inputs that previously (incorrectly) worked.
• Multi-statement or non-SELECT inputs are intentionally blocked; enable write mode only when necessary.

🔄 Upgrade Guidance

1. Pull latest image or reinstall package.
2. Ensure READ_ONLY_MODE env var reflects desired behavior (defaults to secure read-only).
3. Refresh MCP tools in your client to surface updated behavior.

🧩 Commit Summary (since v0.1.4)

Representative commits:

• fix: cross-schema foreign key detection
• feat: query result formatting & related table tests
• docs: improved MCP tool docstrings & usage clarity
• ci: add layered unit → integration workflow
• refactor: wrap_untrusted relocation & unified query assertions
• chore: dependency (sqlparse) + defensive read-only enforcement
• formatter & execution pipeline improvements

Full list: git log v0.1.4..v0.1.5 --oneline

🙏 Thanks

Thanks to contributors enhancing safety, reliability, and developer ergonomics ahead of broader adoption.

---

If legitimate read-only queries are rejected, open an issue with the sanitized SQL and mode settings.

v0.1.4

• New environment variable: ORACLE_CLIENT_LIB_DIR to optionally specify a non-default Oracle client libraries path when using THICK_MODE
• Chinese readme
• Bugfix: Thin mode now works

v0.1.3

• Renamed the project and Docker image to oracle-mcp-server

v0.1.2

• Adds a database connection pool to improve performance

v0.1.1

• Adds thick mode support with THICK_MODE=True environment variable
• The Docker image supports thick mode on Oracle 19c onwards

v0.1.0

Version 0.1.0