# Search Context Improvement Plan
## Current Limitations
Based on the example search session, the main issues are:
1. **Chunk Size Limitations**:
- Code chunks: 1500 chars (often splits functions/classes)
- Config chunks: 1000 chars (may split nested structures)
- AST chunks: 2000 chars (better but still limited)
2. **Limited Context**: Search returns only the matched chunk without surrounding code
3. **No Full-File Option**: Can't easily retrieve complete file context when needed
## Practical Improvements for v0.2.0
### 1. Implement Context Expansion in Search
Add parameters to `search` and `search_code` functions:
```python
def search(
query: str,
n_results: int = 5,
include_context: bool = True, # NEW
context_lines: int = 50, # NEW - lines before/after
...
):
```
Implementation approach:
- When a chunk matches, also fetch adjacent chunks from the same file
- Combine them intelligently to show surrounding context
- Mark the matched section clearly
### 2. Add File-Aware Search Mode
New parameter for searching with file context:
```python
def search_code(
query: str,
file_context_mode: bool = False, # NEW
...
):
```
When enabled:
- Group results by file
- Show file summary with all matching chunks
- Include file-level metadata (imports, classes, functions)
### 3. Increase Default Chunk Sizes
Update defaults based on modern token limits:
```python
# Code Indexer
chunk_size: 1500 → 3000 # Double the context
chunk_overlap: 300 → 600 # Better continuity
# Config Indexer
chunk_size: 1000 → 2500 # Handle larger config sections
# AST Chunker (already good at 2000)
# Keep as is - it's structure-aware
```
### 4. Add Smart Result Expansion
Implement `expand_result()` method that:
- Detects if the chunk is incomplete (e.g., truncated function)
- Automatically includes the complete logical unit
- Uses AST data to find boundaries
### 5. New MCP Tool: get_file_context
Add a dedicated tool for when full file context is needed:
```python
@mcp.tool()
def get_file_context(
file_path: str,
focus_lines: Optional[Tuple[int, int]] = None
) -> Dict[str, Any]:
"""Get full file with optional focus area highlighted"""
```
## Implementation Priority
1. **Quick Win**: Increase chunk sizes (config change)
2. **High Impact**: Add context expansion to search results
3. **Better UX**: Implement get_file_context tool
4. **Advanced**: Smart result expansion using AST data
## Example: Improved Search Flow
Before:
```
1. search("uv requirements") → Limited results
2. grep("uv") → Find files
3. read(file1) → Get context
4. read(file2) → Get more context
```
After:
```
1. search("uv requirements", include_context=True) → Complete context
OR
2. get_file_context("install_global.sh", focus_lines=(60, 70)) → Full file
```
## Configuration
Add to `.env`:
```bash
# Search context settings
SEARCH_CONTEXT_LINES=50
SEARCH_INCLUDE_CONTEXT=true
CODE_CHUNK_SIZE=3000
CONFIG_CHUNK_SIZE=2500
```
## Benefits
1. **Fewer Operations**: 50-70% reduction in follow-up reads
2. **Better Context**: See complete functions/configurations
3. **Faster Workflow**: Get answers in one search
4. **Token Efficiency**: Better information density per operation
## Backward Compatibility
- All changes are additive with sensible defaults
- Existing searches continue to work unchanged
- New parameters are optional
