Skip to content

CLI & Analysis

codemap-search mcp [--no-call-log]
codemap-search codex-config [--server-name NAME]
codemap-search analyze index [--path DIR] [--sort stored|size|lines|symbols|literals|path] [OPTIONS]
codemap-search analyze reads [--path DIR] [--days 1..30] [--sort reads|bytes|size|last|path] [OPTIONS]
codemap-search parse <file>
codemap-search tokenize <ident>
codemap-search codemap [--path P] [--format F]
codemap-search search <query> [-l N]
codemap-search index [dir]
codemap-search benchmark --queries <json> [--dir D]

analyze index inspects the committed index without re-indexing. analyze reads summarizes recorded MCP read, search, and grep calls. Run inside the target repository or pass --path DIR:

Terminal window
codemap-search analyze index --sort size --limit 10
codemap-search analyze index --path /path/to/repo --language rust --filter src/
codemap-search analyze reads --sort bytes --limit 20
codemap-search analyze reads --days 14 --tool search --filter src/
codemap-search analyze reads --view summary --format json
Option Target Description
--path DIR Both Repository directory to inspect
--limit N, -n N Both File rows per page (default 20; 0 returns all)
--offset N Both File rows to skip for pagination
--sort KEY, -s KEY Both index: stored (stored index bytes, default), size (disk file bytes), lines, symbols, literals, path; reads: reads (count, default), bytes (Results bytes), size (disk file bytes), last, path
--order asc|desc Both Ascending for path by default, descending for other keys
--filter TEXT, -f TEXT Both Case-sensitive path substring filter (literal, not glob)
--view summary|files|full Both summary (totals and per-language/tool breakdowns), files (totals and file rows), or full (all sections, default)
--format table|json Both Human-readable tables (default) or compact JSON
--language NAME, -l NAME index Filter by indexed language
--days N, -d N reads Rolling 1–30 day window (default 7)
--tool read|search|grep, -t NAME reads Filter to a single tool
--no-compare reads Skip comparison against the previous window

The MCP analyze tool runs the same query on the active workspace. It defaults to view: "files" and limit: 10 (max 100); pass the returned page.next_offset as offset to fetch the next page:

{"name":"analyze","arguments":{"target":"reads","sort":"bytes","limit":10}}
Section Contents
Index footprint (index) Committed files, segments, disk size, stored index bytes, and call/reference counts
Languages and symbols (index) Files, lines, symbols, exported symbols, literals, and docstrings per language
Files and freshness (index) Indexed files with disk size, symbol/literal counts, and modified/missing status
Current/previous window (reads) Calls, errors, unique files, Reads, Response/Results bytes, and delta vs previous window
Tool and daily activity (reads) Per-tool calls, errors, latency, and UTC daily breakdown
Returned files (reads) Per-file Reads, Results bytes, last seen timestamp, and repeat reads (2+ responses)
  • Reads: One count per file for each successful response that returned source lines or search snippets.
  • Results vs Response: Results measures UTF-8 bytes of delivered file source and snippets; Response measures the full MCP response text, including headings, declarations, and omission notes.

Call metadata is stored for 30 days in .codemap/analysis.sqlite3 without queries or source text. Start the server with codemap-search mcp --no-call-log to disable call logging.