Qartez Mcp

- Pre-computed knowledge graph of the entire codebase
- 43 MCP tools for querying symbols, impact, and structure
- Supports 37 programming languages
- Reduces AI token consumption by ~94% vs. grep-based approaches
- Includes modification guard hooks for Claude Code
- Works with 19 editors and agents out of the box
- Platform support: macOS 13+, Ubuntu 22.04+, Windows, WSL2 (x86_64 and arm64)

Think of these as the standard library for AI code understanding. Each one replaces a multi-step human workflow with a single, token-efficient call the agent can reason about.

Tools are organized into tiers with progressive disclosure. Core tools are always available. Additional tiers can be unlocked on demand via qartez_tools enable: ["analysis"] (or "all").

| Tool | What it does |
|---|---|
| qartez_refs | Trace every usage of a symbol across the codebase, with optional transitive chains. |
| qartez_calls | Call hierarchy: who calls this function, and what does it call. |
| qartez_cochange | Files that historically change together in git. Logical coupling invisible to the import graph. |
| qartez_context | Smart context builder: given files you plan to modify, returns the optimal set of related files to read first. |
| qartez_unused | Dead-code finder: exported symbols with zero importers, pre-materialized at index time. |
| qartez_diff_impact | Batch impact for a git diff range. Pass a revspec like main..HEAD to get changed files with PageRank, union blast radius, convergence points, and co-change omissions. One call replaces N calls to qartez_impact + qartez_cochange. |
| qartez_hotspots | The refactor radar. Ranks files and functions by hotspot score = cyclomatic complexity x PageRank x (1 + churn). Points straight at the highest-risk code in the repo. |
| qartez_clones | Structural code-clone detection via AST shape hashing (identifiers, literals, and comments are normalized away). Finds duplicate logic the human reviewer would never spot. |
| qartez_boundaries | Architecture-boundary enforcement. Declare "these modules may not import those" in .qartez/boundaries.toml and get every violating edge back. suggest=true seeds a starter config from the Leiden clustering. |
| qartez_hierarchy | Type hierarchy queries: find all types implementing a trait/interface, or all traits/interfaces a type implements. Works across Rust, TypeScript, Java, Python, and Go. |
| qartez_trend | Complexity trend over git history: tracks how a function's cyclomatic complexity evolved commit by commit. Flags functions that are GROWING, STABLE, or SHRINKING. |
| qartez_security | Security scanner with 13 built-in rules. Regex-based pattern matching scored by PageRank to prioritize high-impact files. Custom rules via .qartez/security.toml. Filters by severity (low/medium/high/critical) and category. |
| qartez_smells | Code smell detector: finds god functions (high complexity + long body), long parameter lists, and feature envy (methods that use another type more than their own). Tuneable thresholds. |
| qartez_health | Prioritized fix list. Cross-references qartez_hotspots with qartez_smells and buckets files as Critical (hotspot + smell), High (hotspot only), or Medium (smell only). Each entry carries a concrete suggested refactor technique. |
| qartez_refactor_plan | Ordered, safety-annotated refactor plan for one file. Each step names a technique (Extract Method, Introduce Parameter Object), an estimated CC impact category (High/Medium/Low) with a range, and safety signals from impact, test coverage, and caller count. |
| qartez_test_gaps | Test coverage gap analysis via the import graph. Three modes: gaps ranks untested source files by risk, map shows test-to-source mappings, suggest recommends tests to run for a git diff range. |
| qartez_knowledge | Bus-factor analysis. Git-blame-based authorship at file and module level. Surfaces single-author files and modules where knowledge is concentrated in one contributor. |
| qartez_blame | Symbol-scoped git blame. Resolves a function/type name to its line range and blames only those lines: per-hunk commits/authors, or a per-author rollup with each author's latest commit. |
| qartez_semantic | Semantic search using a local embedding model. Natural-language queries ranked by hybrid FTS5 + vector similarity (RRF). Requires the semantic cargo feature and a one-time model download (~270 MB). |
| qartez_path | Shortest call/reference path between two symbols. Forward BFS over the symbol-reference graph returns the ordered symbol/file chain plus a count of alternative shortest paths. Filter edges by kind (call or type). |

| Tool | What it does |
|---|---|
| qartez_rename | Rename a symbol across the entire codebase. Definition, imports, all usages. Preview by default, apply=true to execute. |
| qartez_move | Move a symbol to another file and rewrite all import paths. One MCP call. |
| qartez_rename_file | Rename a file and update every import pointing to it. |
| qartez_replace_symbol | Replace a symbol's whole line range with new source. Caller provides the full replacement including the signature; preview by default, apply=true executes atomically. |
| qartez_insert_before_symbol | Splice new code immediately before an anchor symbol. Anchor lookup goes through the indexed symbol table - no need for the exact surrounding context. |
| qartez_insert_after_symbol | Splice new code immediately after an anchor symbol. Same anchor-based addressing as qartez_insert_before_symbol. |
| qartez_safe_delete | Delete a symbol after reporting every file that still imports it. Refuses to apply when importers exist unless force=true. |

| Tool | What it does |
|---|---|
| qartez_project | Auto-detects your toolchain (Cargo, npm/bun/yarn, Go, Python, Make, Gradle) and runs test/build/lint/typecheck through a single tool. |
| qartez_wiki | Generates a markdown architecture wiki using Leiden community detection on the import graph. Partitions files into clusters, names each one, and emits ARCHITECTURE.md with inter-cluster edges. |
| qartez_workspace | Add or remove workspace domains at runtime. Registers external directories under a custom alias in .qartez/workspace.toml, indexes them, and purges them on removal. |
| qartez_add_root | Register an additional project root at runtime. Indexes the directory, refreshes pagerank/co-change, and hot-attaches a file watcher; the alias is optional (derived from the path basename) and persistence is toggleable for ephemeral roots. |
| qartez_list_roots | List every project root currently tracked by the server with its alias, source (cli/config/runtime), watcher attachment state, file count, and last index timestamp. |
| qartez_maintenance | Inspect and compact .qartez/index.db. Default action stats reports DB / WAL sizes, top tables, current workspace fingerprint, and last full-reindex timestamp. Other actions: checkpoint, optimize_fts (merge FTS5 segments), vacuum_incremental, vacuum, convert_incremental (one-shot conversion to auto_vacuum=INCREMENTAL), purge_stale (drop rows for roots no longer in the workspace). Use this on a bloated DB instead of deleting .qartez/index.db by hand. |

18 tools with complete non-MCP simulations (fair token-to-token comparison):

| Tool | MCP tokens | Without MCP | Savings | Speedup |
|---|---:|---:|---:|---:|
| qartez_cochange | 92 | 14,622 | +99.4% | 2x |
| qartez_context | 107 | 4,489 | +97.6% | 533x |
| qartez_find | 52 | 1,490 | +96.5% | 210x |
| qartez_impact | 352 | 9,243 | +96.2% | 140x |
| qartez_outline | 3,009 | 77,843 | +96.1% | 5x |
| qartez_project | 68 | 1,394 | +95.1% | 0x |
| qartez_unused | 468 | 6,750 | +93.1% | 22x |
| qartez_deps | 166 | 2,286 | +92.7% | 118x |
| qartez_map | 87 | 674 | +87.1% | 1x |
| qartez_rename_file | 27 | 185 | +85.4% | 211x |
| qartez_grep | 127 | 763 | +83.4% | 72x |
| qartez_stats | 155 | 848 | +81.7% | 1x |
| qartez_move | 161 | 701 | +77.0% | 159x |
| qartez_calls | 564 | 2,409 | +76.6% | 3x |
| qartez_refs | 201 | 692 | +71.0% | 26x |
| qartez_read | 150 | 495 | +69.7% | 100x |
| qartez_hierarchy | 735 | 2,056 | +64.3% | 127x |
| qartez_rename | 439 | 648 | +32.3% | 11x |

10 additional analytical tools have no meaningful grep/read equivalent - they solve problems the non-MCP stack cannot solve at all:

qartez_hotspots, qartez_clones, qartez_smells, qartez_test_gaps, qartez_wiki, qartez_boundaries, qartez_trend, qartez_knowledge, qartez_diff_impact, qartez_security.

<details>
<summary>Multi-language bench</summary>

make bench-all runs the same 28-scenario harness against five pinned OSS fixtures - colinhacks/zod (TypeScript), spf13/cobra (Go), encode/httpx (Python), FasterXML/jackson-core (Java), plus the Qartez self-bench (Rust) - then emits a cross-language summary to reports/benchmark-<lang>.md plus a combined matrix. Every tool, every language, every scenario - measured with the cl100k_base tokenizer against a faithful Glob + Grep + Read + git log simulation.

make bench          # Rust self-bench only - fresh measurements
make bench-all      # All 5 languages (Rust, TypeScript, Python, Go, Java) + cross-language summary
make bench-fixtures # Clone and index the pinned fixture repos

Reports land in reports/benchmark.md / reports/benchmark.json for the single-language run, or reports/benchmark-<lang>.md plus a combined cross-language summary for bench-all.

</details>

---