Semantic Search MCP Server

# Playbook: Investigate a Bug ## 1. Reproduce ```bash # Isolate to a fresh data dir to avoid cache interference export SEMANTIC_SEARCH_DATA_DIR=/tmp/codesight-debug-$(date +%s) python -m semantic_search_mcp ``` Use MCP Inspector to send the exact failing tool call. ## 2. Check Logs ```bash # Enable debug logging export LOG_LEVEL=DEBUG python -m semantic_search_mcp ``` ## 3. Inspect State For index corruption issues: ```python import lancedb db = lancedb.connect("~/.semantic-search/data/<repo-hash>/vectors") tbl = db.open_table("chunks") print(tbl.count_rows()) ``` For FTS issues: ```python import sqlite3 conn = sqlite3.connect("~/.semantic-search/data/<repo-hash>/metadata.db") print(conn.execute("SELECT count(*) FROM chunks_fts").fetchone()) ``` ## 4. Verify Read-Only Invariant ```bash # Should find ZERO results — any write to repo_path is a bug grep -rn "open.*'w'" src/semantic_search_mcp/ ``` ## 5. Write a Regression Test Once fixed, add a test in `tests/` that reproduces the bug. File name: `test_<module>_<bug_description>.py`