v0.10.5 - Default Embedding Model Change
SDL-MCP v0.10.5
Important: Default Embedding Model Changed
Action Required for Existing Users: This release changes the default embedding model from all-MiniLM-L6-v2 (384-dim, 256-token) to jina-embeddings-v2-base-code (768-dim, 8192-token).
If you have an existing SDL-MCP installation with semantic search enabled, you need to:
-
Update your config - The config field
embedding.modelnow defaults to"jina-embeddings-v2-base-code". If you explicitly set a model, verify it's still correct. -
Re-index your repositories - Existing embeddings are incompatible (different dimensions). Run:
sdl-mcp index --force
-
Database migration - The schema automatically handles the column changes (
embeddingMiniLM->embeddingJinaCode), but embeddings must be regenerated.
Why the change? Jina's code-specific model provides better code understanding with:
- 768-dimensional vectors (vs 384) for richer representations
- 8192-token context window (vs 256) for longer code spans
- Training specifically on code corpora
What's New
New sdl.file.write Tool
Write non-indexed files (configs, docs, templates) with multiple modes:
- Full content replacement
- Line-range replacement
- Pattern-based replacement
- JSON path updates
- Insert at line / append modes
Workflow Dry-Run Mode
Validate workflow steps and $N references without execution via dryRun: true parameter.
Developer Experience Improvements
Five friction points addressed:
- Symbol ID Opacity:
shortIdfield (first 16 chars) in search results - Action Search: Synonym expansion ("test coverage" finds relevant tools)
- Context Mode Clarity:
contextModeHintexplains precise vs broad mode - Manual Discoverability:
schemaHinttip whenincludeSchemasis false - Truncation Recovery:
cursorfield for easy continuation after truncated windows
CLI Improvements
- Banner display for index/serve commands (HTTP transport)
- Progress display shows current file below progress bar
Bug Fixes
- Windows CI: Allow 8.3 short paths (e.g.,
RUNNER~1) in--repo-pathvalidation - Windows LadybugDB native addon cleanup crash handling
- Embedding column name consistency in schema and tests
- Various test infrastructure improvements
Test Infrastructure Overhaul
- All tests now run isolated for LadybugDB safety
- Improved TAP parsing distinguishes real failures from process exit segfaults
- Test summary with pass/fail counts and segfault tracking
Full Changelog
See CHANGELOG.md for complete details.