GitNexus/gitnexus
azizur100389 131d411ae4
feat(mcp): rank context/impact disambiguation candidates and expose kind/file_path hints (#888)
* feat(mcp): rank context/impact disambiguation candidates and expose kind/file_path hints

The `context` MCP tool already returned `{ status: 'ambiguous', candidates }`
when a name hit multiple symbols, but the candidates were returned in
arbitrary DB order and the only hint it accepted was file_path. The
`impact` tool was worse: when its name resolver found multiple viable
matches it silently picked the first one from a priority UNION, with no
signal back to the caller that a different symbol might have been
intended.

Both failure modes were flagged in issue #470 and reconfirmed in the
comments by a second user who described impact as returning "incorrect
parsing results and meaningless tool calls" in the multi-match case.

Changes:

* Add `resolveSymbolCandidates(repo, query, hints)` private helper on
  LocalBackend. Single place that:
   - Short-circuits on direct uid (zero-ambiguity)
   - Runs the same name-or-qualified-id match as before, with LIMIT 20
     (was 10) so the ranker has headroom instead of arbitrary truncation
   - Preserves the #480 Class/Constructor preference -- when the only
     ambiguity is a Class and its own Constructor, the Class wins
     silently
   - Scores each candidate (pure TS, no extra DB round-trip): base 0.50,
     +0.40 for file_path match, +0.20 for kind match, plus a small
     kind-priority tiebreaker (Class > Interface > Function > Method >
     Constructor) when no explicit kind hint is given
   - Sorts desc by score with stable tiebreakers (shorter filePath,
     then lex uid)
   - Promotes to a single confident resolve when the top score is
     >= 0.95 AND beats the runner-up by >= 0.10 -- lets a strong hint
     cut through without forcing the caller through a disambiguation
     round-trip

* Rewire `context()` to use the shared helper. Response shape is a
  strict superset of today's: candidates gain a `score` field, the
  existing `{ uid, name, kind, filePath, line }` keys are preserved so
  every downstream consumer (rename, eval-server formatter, etc.) keeps
  working. New `kind` input hint accepted.

* Rewire `impact()` to use the shared helper. Now emits the same
  `{ status: 'ambiguous', candidates, impactedCount: 0, risk: 'UNKNOWN' }`
  shape instead of silent first-pick. New inputs accepted:
  `target_uid`, `file_path`, `kind`.

* Update tool schemas in mcp/tools.ts to advertise the new inputs and
  describe ranked disambiguation.

Backward compatibility:

The #480 Class/Constructor collapse is preserved and covered by the
existing java-class-impact integration test (still green). The
ambiguous response shape is a strict superset -- `eval-formatters`
unit test that parses the old shape is unchanged and still passes.
`impact` going from silent-first-pick to structured ambiguous is a
semantic improvement that is the entire point of the issue; callers
relying on silent first-pick now get an actionable response.

Scope declined for v1:

module/community hint -- the issue lists it as one of several hints,
but kind + file_path cover the vast majority of disambiguation needs
in practice, and a community-label filter requires an extra graph
query per candidate. Natural v2 follow-up.

Tests: calltool-dispatch.test.ts gains 5 new cases covering file_path
boost, kind hint boost, impact ambiguous shape, impact target_uid
short-circuit, and score field presence on the existing ambiguous
test. Plus the extended assertions on the existing
`context tool returns disambiguation for multiple matches`.

Verification:
  npx vitest run test/unit/calltool-dispatch.test.ts       -> 64 pass
  npx vitest run test/integration/java-class-impact.test.ts -> pass
  npm run test:unit                                         -> 3642 pass
    (4 pre-existing env failures unchanged: skip-git-cli needs built
    dist/, git-utils tmpdir on Windows worktree -- same on main)
  npx tsc --noEmit                                          -> clean

Closes #470

* fix(mcp): enrich labels from UNION when labels(n)[0] is empty; address review findings

CI on PR #888 caught 13 integration-test failures I did not cover locally:
my resolver refactor collected candidates via `labels(n)[0] AS type`, but
LadybugDB returns an empty string for that projection on certain node
types (most importantly Class). With an empty `type`, impact's downstream
`_runImpactBFS` no longer recognised `symType === 'Class' | 'Interface'`
and stopped seeding Constructor + File nodes into the frontier, so the
"impact(upstream) surfaces the file importer" assertion broke across 11
language fixtures plus 2 OVERRIDES filter tests.

The original impact resolver worked around this by running a prioritised
UNION across Class/Interface/Function/Method/Constructor and picking the
first hit. My refactor dropped that. Fix: keep the simple candidate MATCH
but enrich types afterward via a single scoped UNION query, so every
candidate carries an accurate label for both scoring and downstream
BFS seeding. The UID direct-lookup path is patched the same way.

Also addresses the findings from the senior reviewer on PR #888:

* MIGRATION.md: document the `impact` behavioural change (silent first-
  pick → structured `{ status: 'ambiguous', candidates }`) so downstream
  callers know to branch on `result.status` before reading byDepth/
  summary. `context` is unchanged shape-wise (strict superset).

* New test: `context tool promotes top candidate via scoring when
  multiple rows survive DB pre-filter`. The review flagged that the
  existing file_path test works only because the mock ignores WHERE
  parameters -- the scored-promotion path (top ≥ 0.95 AND gap > 0.09)
  wasn't directly exercised. The new test uses two candidates both in
  App.tsx-containing paths plus a kind hint so promotion is decided by
  scoring, not DB pre-filtering. Also tightened the comment on the
  earlier file_path test to describe the mock vs production divergence
  honestly.

* NIT: added a paragraph explaining why `scored.length >= 2` is kept as
  a defensive guard even though the `normalized.length === 1` early
  return already covers the single-candidate path.

* Integration: two tests in `local-backend-calltool.test.ts` targeted
  `'authenticate'`, which now correctly resolves as ambiguous (two
  Method nodes: AuthService.authenticate and BaseService.authenticate).
  Updated both to pass `file_path: 'src/auth.ts'` so they exercise the
  new disambiguation API and still assert the METHOD_OVERRIDES filtering
  they were originally about.

Edge case fix in the promotion gap check: IEEE754 makes 0.50 + 0.40 +
0.20 - 0.90 = 0.09999999999999998 instead of exactly 0.10, which would
otherwise break the "winner clearly dominates" intent for legitimate
1.00 vs 0.90 cases. Changed `>= 0.10` to `> 0.09`; same user-facing
intent, no floating-point sensitivity.

Verification (all from gitnexus/):
  npx vitest run test/integration/class-impact-all-languages.test.ts
    -> 52 pass (was 11 FAIL on CI before this fix)
  npx vitest run test/integration/local-backend-calltool.test.ts
    -> 18 pass (was 2 FAIL on CI before this fix)
  npx vitest run test/integration/java-class-impact.test.ts
    -> 10 pass (regression guard for #480 preserved)
  npx vitest run test/unit/calltool-dispatch.test.ts
    -> 65 pass (1 new test + 4 from original #470 PR)
  npm run test:unit
    -> 3626 pass, 4 pre-existing env failures unchanged
  npx tsc --noEmit
    -> clean
2026-04-18 12:52:42 +01:00
..
.claude feat: configure prettier with pre-commit hook (#563) 2026-03-28 14:58:04 +00:00
hooks/claude feat: configure prettier with pre-commit hook (#563) 2026-03-28 14:58:04 +00:00
scripts fix: devendor tree-sitter-proto install lifecycle to prevent ENOTEMPTY on global upgrade (#846) 2026-04-15 17:38:47 +01:00
skills feat(hooks): auto-reindex notification with cross-platform hardening (#205) 2026-03-07 08:59:54 +00:00
src feat(mcp): rank context/impact disambiguation candidates and expose kind/file_path hints (#888) 2026-04-18 12:52:42 +01:00
test feat(mcp): rank context/impact disambiguation candidates and expose kind/file_path hints (#888) 2026-04-18 12:52:42 +01:00
vendor fix: devendor tree-sitter-proto install lifecycle to prevent ENOTEMPTY on global upgrade (#846) 2026-04-15 17:38:47 +01:00
.env.example feat: add .env.example, empty batch guard, dimension mismatch warning 2026-03-20 20:15:59 -04:00
.npmignore fix: add preinstall cleanup to prevent ENOTEMPTY on global upgrade (#843) 2026-04-15 14:10:49 +01:00
CHANGELOG.md chore(release): v1.6.2 (#952) 2026-04-18 12:16:51 +01:00
Dockerfile.test feat: add docker support (#848) 2026-04-18 08:39:18 +01:00
package-lock.json chore(deps)(deps): bump glob from 11.1.0 to 13.0.6 in /gitnexus (#867) 2026-04-18 07:40:08 +01:00
package.json chore(release): v1.6.2 (#952) 2026-04-18 12:16:51 +01:00
README.md feat(ci): add release-candidate publish pipeline (#825) 2026-04-14 17:32:47 +01:00
tsconfig.json feat: configure prettier with pre-commit hook (#563) 2026-03-28 14:58:04 +00:00
tsconfig.test.json test: add test suite with vitest (unit + integration + fixtures) 2026-03-01 20:07:02 +05:30
vitest.config.ts feat: configure prettier with pre-commit hook (#563) 2026-03-28 14:58:04 +00:00

GitNexus

Graph-powered code intelligence for AI agents. Index any codebase into a knowledge graph, then query it via MCP or CLI.

Works with Cursor, Claude Code, Codex, Windsurf, Cline, OpenCode, and any MCP-compatible tool.

npm version License: PolyForm Noncommercial


Why?

AI coding tools don't understand your codebase structure. They edit a function without knowing 47 other functions depend on it. GitNexus fixes this by precomputing every dependency, call chain, and relationship into a queryable graph.

Three commands to give your AI agent full codebase awareness.

Quick Start

# Index your repo (run from repo root)
npx gitnexus analyze

That's it. This indexes the codebase, installs agent skills, registers Claude Code hooks, and creates AGENTS.md / CLAUDE.md context files — all in one command.

To configure MCP for your editor, run npx gitnexus setup once — or set it up manually below.

gitnexus setup auto-detects your editors and writes the correct global MCP config. You only need to run it once.

Editor Support

Editor MCP Skills Hooks (auto-augment) Support
Claude Code Yes Yes Yes (PreToolUse) Full
Cursor Yes Yes MCP + Skills
Codex Yes Yes MCP + Skills
Windsurf Yes MCP
OpenCode Yes Yes MCP + Skills

Claude Code gets the deepest integration: MCP tools + agent skills + PreToolUse hooks that automatically enrich grep/glob/bash calls with knowledge graph context.

Community Integrations

Agent Install Source
pi pi install npm:pi-gitnexus pi-gitnexus

MCP Setup (manual)

If you prefer to configure manually instead of using gitnexus setup:

Claude Code (full support — MCP + skills + hooks)

# macOS / Linux
claude mcp add gitnexus -- npx -y gitnexus@latest mcp

# Windows
claude mcp add gitnexus -- cmd /c npx -y gitnexus@latest mcp

Codex (full support — MCP + skills)

codex mcp add gitnexus -- npx -y gitnexus@latest mcp

Cursor / Windsurf

Add to ~/.cursor/mcp.json (global — works for all projects):

{
  "mcpServers": {
    "gitnexus": {
      "command": "npx",
      "args": ["-y", "gitnexus@latest", "mcp"]
    }
  }
}

OpenCode

Add to ~/.config/opencode/config.json:

{
  "mcp": {
    "gitnexus": {
      "command": "npx",
      "args": ["-y", "gitnexus@latest", "mcp"]
    }
  }
}

How It Works

GitNexus builds a complete knowledge graph of your codebase through a multi-phase indexing pipeline:

  1. Structure — Walks the file tree and maps folder/file relationships
  2. Parsing — Extracts functions, classes, methods, and interfaces using Tree-sitter ASTs
  3. Resolution — Resolves imports and function calls across files with language-aware logic
    • Field & Property Type Resolution — Tracks field types across classes and interfaces for deep chain resolution (e.g., user.address.city.getName())
    • Return-Type-Aware Variable Binding — Infers variable types from function return types, enabling accurate call-result binding
  4. Clustering — Groups related symbols into functional communities
  5. Processes — Traces execution flows from entry points through call chains
  6. Search — Builds hybrid search indexes for fast retrieval

The result is a LadybugDB graph database stored locally in .gitnexus/ with full-text search and semantic embeddings.

MCP Tools

Your AI agent gets these tools automatically:

Tool What It Does repo Param
list_repos Discover all indexed repositories
query Process-grouped hybrid search (BM25 + semantic + RRF) Optional
context 360-degree symbol view — categorized refs, process participation Optional
impact Blast radius analysis with depth grouping and confidence Optional
detect_changes Git-diff impact — maps changed lines to affected processes Optional
rename Multi-file coordinated rename with graph + text search Optional
cypher Raw Cypher graph queries Optional

With one indexed repo, the repo param is optional. With multiple, specify which: query({query: "auth", repo: "my-app"}).

MCP Resources

Resource Purpose
gitnexus://repos List all indexed repositories (read first)
gitnexus://repo/{name}/context Codebase stats, staleness check, and available tools
gitnexus://repo/{name}/clusters All functional clusters with cohesion scores
gitnexus://repo/{name}/cluster/{name} Cluster members and details
gitnexus://repo/{name}/processes All execution flows
gitnexus://repo/{name}/process/{name} Full process trace with steps
gitnexus://repo/{name}/schema Graph schema for Cypher queries

MCP Prompts

Prompt What It Does
detect_impact Pre-commit change analysis — scope, affected processes, risk level
generate_map Architecture documentation from the knowledge graph with mermaid diagrams

CLI Commands

gitnexus setup                   # Configure MCP for your editors (one-time)
gitnexus analyze [path]          # Index a repository (or update stale index)
gitnexus analyze --force         # Force full re-index
gitnexus analyze --embeddings    # Enable embedding generation (slower, better search)
gitnexus analyze --skip-agents-md  # Preserve custom AGENTS.md/CLAUDE.md gitnexus section edits
gitnexus analyze --verbose       # Log skipped files when parsers are unavailable
gitnexus mcp                     # Start MCP server (stdio) — serves all indexed repos
gitnexus serve                   # Start local HTTP server (multi-repo) for web UI
gitnexus index                   # Register an existing .gitnexus/ folder into the global registry
gitnexus list                    # List all indexed repositories
gitnexus status                  # Show index status for current repo
gitnexus clean                   # Delete index for current repo
gitnexus clean --all --force     # Delete all indexes
gitnexus wiki [path]             # Generate LLM-powered docs from knowledge graph
gitnexus wiki --model <model>    # Wiki with custom LLM model (default: gpt-4o-mini)

# Repository groups (multi-repo / monorepo service tracking)
gitnexus group create <name>     # Create a repository group
gitnexus group add <name> <repo> # Add a repo to a group
gitnexus group remove <name> <repo> # Remove a repo from a group
gitnexus group list [name]       # List groups, or show one group's config
gitnexus group sync <name>       # Extract contracts and match across repos/services
gitnexus group contracts <name>  # Inspect extracted contracts and cross-links
gitnexus group query <name> <q>  # Search execution flows across all repos in a group
gitnexus group status <name>     # Check staleness of repos in a group

Remote Embeddings

Set these env vars to use a remote OpenAI-compatible /v1/embeddings endpoint instead of the local model:

export GITNEXUS_EMBEDDING_URL=http://your-server:8080/v1
export GITNEXUS_EMBEDDING_MODEL=BAAI/bge-large-en-v1.5
export GITNEXUS_EMBEDDING_DIMS=1024          # optional, default 384
export GITNEXUS_EMBEDDING_API_KEY=your-key   # optional, default: "unused"
gitnexus analyze . --embeddings

Works with Infinity, vLLM, TEI, llama.cpp, Ollama, LM Studio, or OpenAI. When unset, local embeddings are used unchanged.

Multi-Repo Support

GitNexus supports indexing multiple repositories. Each gitnexus analyze registers the repo in a global registry (~/.gitnexus/registry.json). The MCP server serves all indexed repos automatically.

Supported Languages

TypeScript, JavaScript, Python, Java, C, C++, C#, Go, Rust, PHP, Kotlin, Swift, Ruby

Language Feature Matrix

Language Imports Named Bindings Exports Heritage Type Annotations Constructor Inference Config Frameworks Entry Points
TypeScript
JavaScript
Python
Java
Kotlin
C#
Go
Rust
PHP
Ruby
Swift
C
C++

Imports — cross-file import resolution · Named Bindingsimport { X as Y } / re-export tracking · Exports — public/exported symbol detection · Heritage — class inheritance, interfaces, mixins · Type Annotations — explicit type extraction for receiver resolution · Constructor Inference — infer receiver type from constructor calls (self/this resolution included for all languages) · Config — language toolchain config parsing (tsconfig, go.mod, etc.) · Frameworks — AST-based framework pattern detection · Entry Points — entry point scoring heuristics

Agent Skills

GitNexus ships with skill files that teach AI agents how to use the tools effectively:

  • Exploring — Navigate unfamiliar code using the knowledge graph
  • Debugging — Trace bugs through call chains
  • Impact Analysis — Analyze blast radius before changes
  • Refactoring — Plan safe refactors using dependency mapping

Installed automatically by both gitnexus analyze (per-repo) and gitnexus setup (global).

Requirements

  • Node.js >= 18
  • Git repository (uses git for commit tracking)

Release candidates

Stable releases publish to the default latest dist-tag. When a pull request with non-documentation changes merges into main, an automated workflow also publishes a prerelease build under the rc dist-tag, so early adopters can try in-flight fixes without waiting for the next stable cut. (Docs-only merges are skipped.)

# Try the latest release candidate (pre-stable — may change at any time)
npm install -g gitnexus@rc
# — or —
npx gitnexus@rc analyze

Release-candidate versions follow the standard semver prerelease format X.Y.Z-rc.N, where X.Y.Z is the next stable target (bumped from the current latest by patch by default; minor or major when kicking off a bigger cycle) and N increments per published rc. Example sequence: 1.6.2-rc.1, 1.6.2-rc.2, …, then once 1.6.2 ships stable, 1.6.3-rc.1. See the Releases page for the full list; stable latest is unaffected.

Troubleshooting

Cannot destructure property 'package' of 'node.target' as it is null

This crash was caused by a dependency URL format that is incompatible with certain npm/arborist versions (npm/cli#8126). It is fixed in gitnexus v1.6.2+. Upgrade to the latest version:

npx gitnexus@latest analyze          # always uses the newest release
# — or —
npm install -g gitnexus@latest       # upgrade a global install

If you still hit npm install issues after upgrading, these generic workarounds may help:

npm install -g npm@latest            # update npm itself
npm cache clean --force              # clear a possibly corrupt cache

Installation fails with native module errors

Some optional language grammars (Dart, Kotlin, Swift) require native compilation. If they fail, GitNexus still works — those languages will be skipped.

If npm install -g gitnexus fails on native modules:

# Ensure build tools are available (Linux/macOS)
# Ubuntu/Debian: sudo apt install python3 make g++
# macOS: xcode-select --install

# Retry installation
npm install -g gitnexus

Analysis runs out of memory

For very large repositories:

# Increase Node.js heap size
NODE_OPTIONS="--max-old-space-size=16384" npx gitnexus analyze

# Exclude large directories
echo "vendor/" >> .gitnexusignore
echo "dist/" >> .gitnexusignore

Privacy

  • All processing happens locally on your machine
  • No code is sent to any server
  • Index stored in .gitnexus/ inside your repo (gitignored)
  • Global registry at ~/.gitnexus/ stores only paths and metadata

Web UI

GitNexus also has a browser-based UI at gitnexus.vercel.app — 100% client-side, your code never leaves the browser.

Local Backend Mode: Run gitnexus serve and open the web UI locally — it auto-detects the server and shows all your indexed repos, with full AI chat support. No need to re-upload or re-index. The agent's tools (Cypher queries, search, code navigation) route through the backend HTTP API automatically.

License

PolyForm Noncommercial 1.0.0

Free for non-commercial use. Contact for commercial licensing.