mirror of https://github.com/DeusData/codebase-memory-mcp.git synced 2026-11-10 09:13:53 +08:00

Files

T

Martin Vogel 02bc3b7352 Release v0.3.0: 25 languages, install/update CLI, skills, benchmark

Language expansion (13 → 25):
  Add Ruby, C, Bash, Zig, Elixir, Haskell, OCaml, HTML, CSS, YAML,
  TOML, HCL with tree-sitter grammars and language specs.
  Restore Erlang and SQL with extraction fixes.

CLI install/update/uninstall commands:
  Auto-detect Claude Code and Codex CLI, register MCP server,
  install task-specific skills, self-update with SHA-256 verification.

Embedded skills (4):
  exploring, tracing, quality, reference — auto-trigger in Claude Code
  for graph-first code discovery.

Pipeline improvements:
  Docstring extraction, enrichment pass for params/returns/complexity,
  graceful context cancellation, improved test detection.

35-language benchmark (BENCHMARK.md):
  12 questions × 35 repos, 91.8% overall score, Linux kernel stress
  test (20K nodes, zero timeouts). Replaces old benchmark artifacts.

2026-03-01 18:01:52 +01:00

1.7 KiB

Raw Blame History

Codebase Knowledge Graph

This project has a code knowledge graph available via MCP tools. Use it for structural questions instead of grep — one graph query returns what would take dozens of file-by-file searches.

Finding Code

"Who calls X?" → use trace_call_path with direction=inbound
"What does X call?" → use trace_call_path with direction=outbound
"Find functions matching pattern" → use search_graph with name_pattern
"Find all routes" → use search_graph with label=Route
"Show codebase structure" → use get_graph_schema for overview

Tracing Dependencies

Always discover exact names first: search_graph(name_pattern=".*Partial.*")
Then trace: trace_call_path(function_name="ExactName", direction="both")
Cross-service HTTP calls: query_graph("MATCH (a)-[r:HTTP_CALLS]->(b) RETURN a.name, b.name, r.url_path")
Read source: get_code_snippet(qualified_name="project.path.FunctionName")

Quality Analysis

Dead code: search_graph with max_degree=0, exclude_entry_points=true, relationship=CALLS, direction=inbound
High fan-out: search_graph with min_degree=10, relationship=CALLS, direction=outbound
Change coupling: query_graph("MATCH (a)-[r:FILE_CHANGES_WITH]->(b) WHERE r.coupling_score >= 0.5 RETURN ...")

Important

Always check list_projects first — run index_repository if the project is missing
Use search_graph to discover exact names before trace_call_path (it requires exact match)
The graph doesn't index text content — use grep for string literals, error messages, config values
Results default to 10 per page — check has_more and use offset to paginate
Use project parameter when multiple repos are indexed

1.7 KiB Raw Blame History