v2.1.1 - Schema-Grounded CAG & Runtime Accuracy
Strict schema-first generation, safer cache scoping, and repair-before-execute
Release Date: March 17, 2026
Author: Arjun Trivedi
Email: arjuntrivedi42@yahoo.com
Repository: https://github.com/4R9UN/mcp-kql-server
🚀 What's New in v2.1.1
This release focuses on correctness and schema fidelity. The server now prefers cached schema/CAG context over re-indexing, generates KQL from ranked schema context, and can repair invalid client-generated KQL only when a real schema proves the replacement.
1. Stricter CAG and Schema Memory Usage
- Schema-First Generation: NL2KQL generation now uses ranked table and column context from schema memory before building candidate queries.
- Strict Table Context:
schema_memory(operation="get_context")can now be scoped to a specific table and returns:- allowed columns
- recommended columns
- preferred time column
- compact CAG context
- No Redundant Reindexing: Existing cached schema is reused before live discovery, and empty placeholder discovery paths no longer overwrite rich table schemas.
2. Runtime Query Accuracy
- Schema-Grounded Repair: When direct KQL contains invalid columns, the server now attempts a schema-backed repair and only executes the repaired query if it fully revalidates.
- Improved Join-Aware Generation: Candidate generation now includes join-aware paths when join hints exist in memory.
- Generation Telemetry: Memory stats now include generation event counts and candidate metrics for tuning.
3. Cache and Runtime Stability
- Scoped Query Cache: Cached results are now isolated by query text, cluster, database, and cache namespace.
- Safe Cache Clearing:
schema_memory(operation="clear_cache")now clears real cached state rather than only returning a success message. - Shared Connection Pooling: Execution paths now use the shared pool instead of a duplicate local client cache path.
- Safer Health Checks: Pool health checks probe only with a registered database and avoid false recycling when no database context exists.
4. Repository Cleanup
- Removed Scratch Docs: Deleted ad hoc analysis documents that were not part of the shipped project docs.
- Version Alignment: Package metadata, docs, server manifest, and tests are aligned to
2.1.1.
🐛 Bug Fixes
- Fixed repeated schema writes when schema was already available in memory/CAG.
- Prevented empty placeholder schemas from overwriting real indexed schemas.
- Fixed direct-query failures caused by invalid generated/client column names such as non-schema time aliases.
📦 Installation & Upgrade
New Installation
pip install mcp-kql-server==2.1.1Upgrade from Previous Versions
pip install --upgrade mcp-kql-server✅ Quality Assurance
- Tests: 76 passed, 1 skipped
- Verified: Generation, execution, runtime cache, health-check, and package-version regression suites passed