에이전트 강화
게시자의 자체 명령입니다. PlugMCP 엔드포인트가 아닙니다.
Host a copy with our keys and logs.
uvx n24q02m-crg==3.32.0Claude Code에는 로컬 명령용 설치 링크가 없습니다. 아래 명령을 복사해 터미널에서 실행합니다.
EMBEDDING_MODELS=<value> LOCAL_EMBEDDING_MODEL=<value> LOCAL_RERANK_MODEL=<value> LOCAL_EMBEDDING_DIM=<value> LOCAL_EMBEDDING_MODEL_FILE=<value> LOCAL_EMBEDDING_POOLING=<value> LOCAL_EMBEDDING_NORMALIZE=<value> HULL_EMBED_API_KEY=<secret> HULL_CHAT_API_KEY=<secret> EMBEDDING_BACKEND=<value> claude mcp add io.github.n24q02m-crg -- uvx n24q02m-crg==3.32.0터미널에서 실행합니다.
{ "mcpServers": { "io.github.n24q02m-crg": { "command": "uvx", "args": [ "n24q02m-crg==3.32.0" ], "env": { "EMBEDDING_MODELS": "<value>", "LOCAL_EMBEDDING_MODEL": "<value>", "LOCAL_RERANK_MODEL": "<value>", "LOCAL_EMBEDDING_DIM": "<value>", "LOCAL_EMBEDDING_MODEL_FILE": "<value>", "LOCAL_EMBEDDING_POOLING": "<value>", "LOCAL_EMBEDDING_NORMALIZE": "<value>", "HULL_EMBED_API_KEY": "<secret>", "HULL_CHAT_API_KEY": "<secret>", "EMBEDDING_BACKEND": "<value>" } } }}claude_desktop_config.json
원격 URL인가요, 내 컴퓨터에서 실행하는 명령인가요? 두 설치 방식의 차이
PlugMCP에서 실행
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
실행게시자 제공
Renamed (2026-09-13): repo is now
crg— CLI-first (crgcommand). PyPI package staysbetter-code-review-graph; MCP server is a secondary surface.
mcp-name: io.github.n24q02m/crg
Knowledge graph for token-efficient code reviews -- semantic search and call-graph resolution across your codebase.
Token-efficient code review knowledge graph: semantic search and call-graph resolution.
An MCP server that parses your codebase with Tree-sitter, builds a structural graph of functions/classes/imports, and gives Claude (or any MCP client) precise context so it reads only what matters instead of the whole tree. Semantic search runs through the local ONNX model registry from fastretrieval by default (zero config, no API key), with an optional cloud embedding chain. Fork of code-review-graph with fixed multi-word search, qualified call resolution, dual-mode embeddings, output pagination, and production CI/CD.
v2.0 adds temporal columns (valid_from_sha / valid_to_sha on every node + edge) and an opt-in security scanner. The schema migration is auto-applied on first GraphStore open, and a backup of the pre-2.0 DB is saved to <graph_db>.pre-2.0.bak so you can roll back. See BREAKING_CHANGES.md for the full schema-change list, behavior changes, environment requirements, and the downgrade procedure (CRG_DOWNGRADE_TO_1_X=1 uv run crg).
For OMP and other local coding harnesses, the primary surface is the package CLI
plus the bundled skills/ workflows. The skills invoke the CLI directly and do
not require an MCP server mapping.
# Run without a persistent install (short `crg` script; PyPI package name stays
# better-code-review-graph, so `uvx` needs the explicit --from form)
uvx --python 3.13 --from better-code-review-graph crg graph build --full-rebuild \
--repo-root /path/to/repo
uvx --python 3.13 --from better-code-review-graph crg graph stats \
--repo-root /path/to/repo
# Or install the console scripts (installs both `crg` and the legacy long name)
pip install better-code-review-graph
crg query search --search-query "authentication" \
--repo-root /path/to/repo
The optional Semgrep engine for deeper security scans is a separate extra:
pip install 'better-code-review-graph[security]'
MCP stdio remains a secondary protocol adapter for clients that require it:
{
"mcpServers": {
"crg": {
"command": "uvx",
"args": ["--python", "3.13", "better-code-review-graph"],
"env": { "MCP_TRANSPORT": "stdio" }
}
}
}
Install matrix (stdio unless noted; the CLI-first usage above stays the primary surface):
Install with an AI agent -- paste this to your AI coding agent:
Install MCP server
crgfollowing the steps at https://raw.githubusercontent.com/n24q02m/claude-plugins/main/plugins/better-code-review-graph/setup-with-agent.md
Full CLI usage is in CLI. Optional per-client MCP setup is at mcp.n24q02m.com/servers/better-code-review-graph/setup/.
Two ways to run the server, plus the surfaces to consume it.
# instance config lives at $CRG_CONFIG_DIR or ~/.crg (same schema as hull's
# config.toml); a default no-auth template is fine for local dev
uv run better-code-review-graph server start
# -> http://127.0.0.1:8000/mcp (no-auth is refused on any off-loopback bind)
cp docker-config/config.example.toml docker-config/config.toml
# mint a token + hash (hull CLI ships with crg's dependency tree):
# TOKEN=$(python -c "import secrets; print(secrets.token_urlsafe(32))")
# hull token hash "$TOKEN" # paste output into token_hash in config.toml
docker compose up -d
The compose file builds the Dockerfile's http target (image crg:http),
publishes the container on host loopback only
(127.0.0.1:${CRG_PORT:-8772}:8080), and mounts
./docker-config/config.toml read-only at the container's config dir
(CRG_CONFIG_DIR=/data/config); graph state persists in the crg-data
named volume. All state stays on your machine.
CLI (local graphs, no server needed):
crg graph build --full-rebuild --repo-root /path/to/repo
crg query search --search-query "authentication" --repo-root /path/to/repo
MCP over HTTP (remote/always-on): point any MCP client at
http://127.0.0.1:8772/mcp with Authorization: Bearer <token>:
{
"mcpServers": {
"better-code-review-graph": {
"url": "http://127.0.0.1:8772/mcp",
"headers": { "Authorization": "Bearer <token>" }
}
}
}
Each task (embed / rerank / chat / jev_score) resolves its own OpenAI-spec
provider cell from [models.<task>] in the instance config: an independent
base_url + api_key + model. Cloud or local is purely a config choice —
point base_url at OpenRouter, a vendor, or your own local gateway
(e.g. http://host.docker.internal:11434/v1). API keys are host-only
material: keep them in config.toml (gitignored under docker-config/) or
supply via HULL_<TASK>_API_KEY env; they are never end-user supplied.
CRG is local-first for coding workflows:
<repo>/.crg/graph.db unless an explicit
multi-user/self-host configuration selects another data directory.The repo ships a smithery.yaml so the server can be built and
run through Smithery. It deploys over stdio and needs
no startup configuration -- the config schema is empty, and any optional cloud
embedding/summary keys are supplied at runtime through the server's own config
flow (see Configuration below). The launch command is the same
uvx invocation as a local install:
startCommand:
type: stdio
commandFunction: |-
(config) => ({ command: 'uvx', args: ['--python', '3.13', 'crg'] })
Everything works out of the box with zero configuration -- semantic search
uses the local ONNX registry from fastretrieval
(Qwen3-Embedding-0.6B is the current built-in reference entry, ~570 MB
downloaded on first graph embed). This reference entry is not a Qwen-only
boundary: any built-in registry ID or valid non-Qwen artifact manifest follows
the same resolver. All environment variables below are optional and only needed
for cloud embeddings, LLM summaries, or an explicit BYO local artifact.
Embeddings select the first provider/model entry in EMBEDDING_MODELS; later
entries are retained as configuration but are not runtime fallbacks. The summary
model comes from the model field of the [models.chat] cell (an unconfigured
cell disables summaries). Transport always comes from the per-task
[models.<task>] cell (OpenRouter pre-wired default) — model-name prefixes
only select wire details, never keys.
Cohere embed-v4.0 requests and stores 1024 dimensions; other backends retain
768-dimensional storage. CRG never slices, pads, or silently accepts a different
provider width. The embedding row's model and byte width must match before reuse.
Run graph(action="embed") after changing models or upgrading an old 768-wide
Cohere index. Searches reject incompatible widths before a provider call; graph
nodes are retained and re-embedding replaces only stale vectors.
The key lives in the per-task cell's api_key (host-only; see the table below).
Keys alone never select models: an empty embedding chain stays local, and an empty summary
chain stays disabled. A configured cloud error does not fall back to local or
another provider. Summarizers require a chat-completion model.
Vendor-specific key env vars (JINA_AI_API_KEY, GEMINI_API_KEY,
OPENAI_API_KEY, GOOGLE_VERTEX_EXPRESS_API_KEY, …) are not read by
crg; the removed per-vendor dispatch lived in the pre-de-host stack.
When LOCAL_RERANK_MODEL is configured, semantic vector search retrieves a
bounded candidate pool of min(max(limit * 4, limit), 100) rows, applies the
existing kind, repo, and live-row filters, then reranks that pool and returns
at most limit rows. The response uses search_mode="semantic_reranked" and
adds rerank_score while preserving similarity_score. Blank keeps the
existing limit * 2 vector path and search_mode="semantic". Configured
reranker failures return an explicit error; CRG does not silently fall back to
vector or keyword results. Keyword searches, including as_of snapshots, do
not invoke the reranker.
~/.crg/config.toml (write by hand, or start from hull config init which
pre-wires OpenRouter defaults):
[models.embed]
base_url = "https://openrouter.ai/api/v1"
api_key = "sk-or-..."
model = "cohere/embed-v4.0"
[models.chat]
base_url = "https://openrouter.ai/api/v1"
api_key = "sk-or-..."
model = "minimax/minimax-m3:free"
Equivalent env surface for the MCP server entry (keys override the cells;
model names still come from EMBEDDING_MODELS / the model field):
{
"mcpServers": {
"crg": {
"command": "uvx",
"args": ["--python", "3.13", "better-code-review-graph"],
"env": {
"MCP_TRANSPORT": "stdio",
"EMBEDDING_MODELS": "cohere/embed-v4.0",
"HULL_EMBED_API_KEY": "sk-or-...",
"HULL_CHAT_API_KEY": "sk-or-..."
}
}
}
}
Cohere embedding is paid. Authorize a bounded budget before a live index/query;
the Minimax-free completion choice does not make embeddings free. Model-name
prefixes (cohere/…, openrouter/…) select wire details only — transport and
credentials come from the [models.*] cells, which the host owns.
CRG currently has no cloud rerank call: LOCAL_RERANK_MODEL is its only
reranking path. Setting RERANK_MODELS or RERANK_API_BASE does not enable one.
Six tools, each grouping related actions to keep the tool surface small.
graph -- Graph lifecycleActions: build | update | stats | embed | export | summarize
query -- Graph queriesActions: query | search | impact | large_functions | spot_check | renamed_in_diff | diff
Most read actions accept as_of=<sha> for temporal (point-in-time) snapshots
and repo=<repo_id> to scope a federated multi-repo graph.
review -- Code review contextActions: context (default) | delta
Token-optimized review context with structural summary, impacted nodes, source
snippets, and review guidance. context auto-detects changed files from the
git diff; delta (with from_sha/to_sha, optional show_line_shifts)
surfaces refactor moves between two commits.
config -- Server configuration and credential setupActions: status | set | cache_clear | setup_status | setup_start | setup_skip | setup_reset | setup_complete
security -- Security scanningActions: scan | report | suppress | rule_list
The semgrep engine requires the [security] extra and runs Semgrep's
p/auto registry pack plus a 3-rule curated overlay.
help -- Full documentationTopics: graph | query | review | config | security | recipes
Returns complete documentation for each tool. Use when the compressed descriptions above are insufficient.
The package installs two console scripts: crg (primary) and
crg (legacy long name). Running either with no
arguments starts the MCP server over stdio; a leading positional argument
routes to a local CLI subcommand that calls the same domain services used by
the MCP adapter. Run them directly after pip install, or without a
persistent install via uvx --python 3.13 --from better-code-review-graph crg ....
# Start the MCP server over stdio (default -- no subcommand)
crg
# Build, inspect, and embed the local graph
crg graph build
crg graph stats
crg graph embed
# Query relationships and impact
crg query query \
--pattern callers_of --target "path/to/module.py::function"
crg query search --search-query "authentication"
crg query impact --changed-files src/app.py
# Produce review context and run a local security scan
crg review context --base HEAD~1
crg security scan --engine heuristic
CLI subcommands print structured JSON and exit non-zero on an error.
What this fork fixes versus the upstream code-review-graph:
How crg stacks up against direct competitors in each pillar:
Sources: Greptile · Greptile pricing · Sourcegraph MCP · CodeGraph. Cells marked ? are capabilities the competitor does not publicly document, not confirmed absences.
:ro (read-only).[models.*] cell base_url values are validated before any outbound call.To report a vulnerability, see SECURITY.md.
git clone https://github.com/n24q02m/crg
cd crg
uv sync --group dev
uv run pytest
uv run crg
Requirements: Python 3.13, uv.
This plugin implements TC-Local (machine-bound, single trust principal). See the mcp-core trust model for full classification.
Graph, security scan cache, and suppression state now use the package-owned
.crg/ directory. Run graph(action="build", full_rebuild=true)
once after upgrading, followed by graph(action="embed") if semantic search is
needed. The old .better-code-review-graph/ state directory plus the ambiguous
.code-review-graph/ and .code-review-graph.db paths and their SQLite
sidecars are left untouched, not migrated. Review and reapply any desired
suppression rules explicitly.
The v2.0 release added temporal columns (valid_from_sha / valid_to_sha
on every node and edge) plus an opt-in security scanner. The schema migration
is auto-applied on first GraphStore open, and a backup of the pre-2.0 DB is
written to <graph_db>.pre-2.0.bak. To downgrade and restore it:
CRG_DOWNGRADE_TO_1_X=1 uvx better-code-review-graph
Full schema-change list, behavior changes, and rollback procedure: BREAKING_CHANGES.md. Release-by-release history: CHANGELOG.md.
Full docs at mcp.n24q02m.com/servers/better-code-review-graph/setup/:
Use the help tool from any MCP client for inline per-tool reference.
Apache-2.0 -- See LICENSE.
Token-efficient code review knowledge graph: semantic search and call-graph resolution.
{ "mcpServers": { "io.github.n24q02m-crg": { "command": "uvx", "args": [ "n24q02m-crg==3.32.0" ], "env": { "EMBEDDING_MODELS": "<value>", "LOCAL_EMBEDDING_MODEL": "<value>", "LOCAL_RERANK_MODEL": "<value>", "LOCAL_EMBEDDING_DIM": "<value>", "LOCAL_EMBEDDING_MODEL_FILE": "<value>", "LOCAL_EMBEDDING_POOLING": "<value>", "LOCAL_EMBEDDING_NORMALIZE": "<value>", "HULL_EMBED_API_KEY": "<secret>", "HULL_CHAT_API_KEY": "<secret>", "EMBEDDING_BACKEND": "<value>" } } }}도구
아직 도구 목록이 없습니다. 원격 서버는 첫 상태 확인 후에 도구 목록이 생깁니다.
FAQ
이 페이지에서 Cursor, Claude Desktop, Claude Code 또는 VS Code용 stdio 설정을 복사합니다. PlugMCP 프록시가 아니라 게시자의 설치 방식을 사용합니다.
게시자가 등록한 환경 변수: EMBEDDING_MODELS, LOCAL_EMBEDDING_MODEL, LOCAL_RERANK_MODEL, LOCAL_EMBEDDING_DIM, LOCAL_EMBEDDING_MODEL_FILE, LOCAL_EMBEDDING_POOLING, LOCAL_EMBEDDING_NORMALIZE, HULL_EMBED_API_KEY, HULL_CHAT_API_KEY, EMBEDDING_BACKEND.
Crg은(는) stdio를 통해 로컬에서 실행됩니다.
마지막 확인: 2026-11-13. Token-efficient code review knowledge graph: semantic search and call-graph resolution.