By iliaal
Claude Code integration for cocoindex-code (ccc). Slash commands for onboarding, reset, and re-index; git hook installer for automatic incremental indexing; standard exclude presets that avoid common retrieval noise (tests, plans, compiled views, ML embedding data).
Run ccc retrieval benchmarks on all corpora, save timestamped scorecards, print summary
Onboard a project to cocoindex-code (init, excludes, index, MCP, hooks, hint)
Force an incremental ccc re-index of a project (use when git hooks aren't firing or after a bulk change)
Reset and fully re-build a project's ccc index (use after settings changes or device switches)
Claude Code plugin for cocoindex-code integration: semantic retrieval via MCP, automatic index freshness via git hooks, and one-command project onboarding.
/ccc-onboard <path> — initialize ccc on a project, apply standard excludes, build the index, register the MCP server with Claude Code, install git hooks for automatic refresh, and write a local .claude/CLAUDE.md hint file. Single command, six idempotent steps./ccc-reset <path> — drop and rebuild a project's ccc index. Use after settings changes, device switches (CPU ↔ CUDA), or index corruption. Destructive, requires confirmation./ccc-reindex <path> — force an incremental re-index. Use when you want an immediate refresh without waiting for a git hook.bin/ccc-install-hook <repo> <ccc-root> — standalone installer for post-commit, post-merge, post-checkout hooks that fire a non-blocking ccc index in the background. Called automatically by ccc-onboard.cocoindex-code is a capable semantic retrieval engine but ships as a standalone CLI. Integrating it into a Claude Code workflow on a real codebase requires five separate steps (init, exclude tuning, initial index, MCP registration with a sh -c "cd PATH && exec ccc mcp" wrapper because ccc mcp refuses to start outside an initialized project, git hooks for freshness). This plugin wraps all five into one command and adds freshness automation.
Background and benchmark methodology: ~/ai/wiki/tools/semantic-retrieval-test-noise.md.
The ccc-onboard script applies these exclusions out of the box (on top of cocoindex-code's defaults):
**/tests/**, **/test/**, **/*Test.php, **/*.test.ts, **/*.spec.ts, **/test_*.py, **/*_test.py**/docs/plans/**, **/plans/**, **/docs/****/package-lock.json, **/composer.lock, **/Cargo.lock, **/yarn.lock**/.terraform**/.venv, **/venv, **/env, **/__pycache__, **/.pytest_cache, **/coverage**/*.min.js, **/*.min.css**/storage/framework/views/****/*embeddings*.json, **/similarity_matrix*.json, **/test_outputs/**The test exclusion is the important one. Measured on a 14-case php-src benchmark: including tests in the index drops recall@10 from 1.00 to 0.79. Tests are written in symptom language ("crash when X happens") which semantically matches bug-report queries better than the actual source file's mechanism-language. If you include tests, they crowd out the implementation in top-N results. See the wiki entry for the full write-up.
# Clone this repo
git clone https://github.com/<you>/ccc-plugin.git ~/ai/ccc-plugin
# Register the marketplace with Claude Code
claude plugin marketplace add ~/ai/ccc-plugin
# Install the plugin
claude plugin install ccc-tools@ccc-plugin
Prerequisites:
cocoindex-code installed (uv tool install --upgrade cocoindex-code --prerelease explicit --with "cocoindex>=1.0.0a24")ccc doctor reports greenInside a Claude Code session:
/ccc-onboard ~/my-project
The agent runs the shell script, reports the six-step result, verifies MCP connectivity, runs a sanity query, and summarizes the state. After that, ~/my-project is queryable via the ccc-my-project MCP tool, the index auto-refreshes on commits, and there's a local hint file at ~/my-project/.claude/CLAUDE.md teaching future Claude Code sessions when to prefer semantic search over grep.
CCC_BENCH_CORPUS_DIR — directory where /ccc-bench looks for *-eval.yaml corpora and writes timestamped scorecards to history/. Defaults to $HOME/ai/retrieval-bench. Set this if you keep your benchmark corpora somewhere else. The --corpus-dir CLI flag overrides both the env var and the default..
├── .claude-plugin/marketplace.json # marketplace manifest
└── plugins/ccc-tools/
├── .claude-plugin/plugin.json # plugin manifest
├── commands/
│ ├── ccc-onboard.md
│ ├── ccc-reset.md
│ └── ccc-reindex.md
└── bin/
├── ccc-onboard # bash, ~130 lines
└── ccc-install-hook # bash, ~45 lines
Commands reference their bash counterparts via ${CLAUDE_PLUGIN_ROOT}/bin/..., so the plugin is self-contained and relocatable.
MIT. See LICENSE.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub iliaal/ccc-plugin --plugin ccc-toolsClaude Code integration for CodeSage, a Rust-based code intelligence engine. Provides slash commands for onboarding a project (init + index + global MCP registration + git hooks + agent hint), reset, incremental reindex with model-mismatch cleanup, and retrieval benchmarking. A single global MCP server serves every onboarded project, routed by an absolute `project` argument on each call.
AST-based semantic code search via the ccc CLI. Bundles the ccc skill so coding agents handle init, indexing, and search automatically.
Lightning-fast semantic code search with auto-indexing. Automatically manages sgrep watch during Claude sessions.
Precise local semantic code search via MCP. Indexes your codebase with Go AST parsing, embeds with Ollama or LM Studio, and exposes vector search to Claude through an MCP server — no cloud, no npm.
Optimized file search, semantic indexing, and persistent memory for Claude Code — with optional sync to a self-hosted web dashboard
Structural codebase indexing for efficient navigation. Reduces token consumption by 60-80% through targeted line-range reads instead of full file scans.
Semantic code search for Claude Code. Automatically indexes your project and provides intelligent search capabilities.