better-code-review-graph

Embeddings and summaries are each driven by anordered model chain-- a CSV ofprovider/modelentries where the order is the litellm fallback order (first entry is the active model). The provider is inferred from the model prefix, so the matching<PROVIDER>_API_KEYis all you need to add.

All vectors are stored at a fixed 768 dimensions (MRL truncation), so the embeddings table schema stays valid across providers. Switching embeddingmodelchanges the vector space; embeddings are tracked per provider and a provider switch triggers re-embedding rather than mixing incomparable vectors.

Cloud models need the provider key for whatever prefixes appear in your chains. Without any cloud key the server stays on local ONNX. Summarizers must expose a chat-completion API (so Jina and Cohere are embedding-only).

CRG intentionally exposes no local reranker settings because this server has no local reranker path. A custom external embedding ID without a manifest must provideLOCAL_EMBEDDING_DIM; a local artifact directory must provide a validfastretrieval-manifest.json, otherwise startup fails closed.

The relay setup form has an optionalworkspace usernamefield. Entering the same username always lands you in the same per-subbucket, so your keys and graph stay reachable across a re-authorization and across devices, instead of being tied to the one-off subject minted for each/authorizeround-trip. Leaving it blank keeps the previous per-authorize behaviour.

Trust boundary: when the form is gated by asharedMCP_RELAY_PASSWORD, the username is a partition key, not a secret -- anyone who knows that password can type any username and reach that bucket. That is fine for a trusted group; an untrusted multi-tenant deployment needs a per-user secret or delegated OAuth instead.

One-time migration:existing users must re-enter their credentials once after this change. Nothing is deleted; credentials stored under the old random subject are simply no longer addressed.

Seven tools, each grouping related actions to keep the tool surface small.

Actions:build|update|stats|embed|export|summarize

Actions:query|search|impact|large_functions|spot_check|renamed_in_diff|diff

Most read actions acceptas_of=<sha>for temporal (point-in-time) snapshots andrepo=<repo_id>to scope a federated multi-repo graph.

Token-optimized review context with structural summary, impacted nodes, source snippets, and review guidance.contextauto-detects changed files from the git diff;delta(withfrom_sha/to_sha, optionalshow_line_shifts) surfaces refactor moves between two commits.

config-- Server configuration and credential setup

Actions:status|set|cache_clear|setup_status|setup_start|setup_skip|setup_reset|setup_complete

Actions:scan|report|suppress|rule_list

Runningbetter-code-review-graphwithno argumentsstarts the MCP server over stdio (this is what an MCP client launches). A leading positional argument routes to a subcommand instead -- handy for building or embedding the graph directly from a shell or CI step, before any MCP client connects. Run these withuvx(oruv runfrom a source checkout):

# Start the MCP server over stdio (default -- no subcommand) uvx better-code-review-graph # Build (or incrementally update) the graph for the current repo uvx better-code-review-graph graph build # Full re-parse of every file instead of a git-diff incremental uvx better-code-review-graph graph build --full-rebuild # Compute embeddings for semantic search (local ONNX by default) uvx better-code-review-graph graph embed

How better-code-review-graph stacks up against direct competitors in each pillar:

- Graceful fallbacks-- Cloud embedding failure falls back to local ONNX.
- Error handling-- Tools return error strings with fix suggestions, never crash.
- Read-only mount-- Docker mode mounts the repo as:ro(read-only).
- SSRF-guarded endpoints-- CustomEMBEDDING_API_BASE/LLM_API_BASEURLs are validated before any outbound call.

git clone https://github.com/n24q02m/better-code-review-graph cd better-code-review-graph uv sync --group dev uv run pytest uv run better-code-review-graph

The v2.0 release addedtemporal columns(valid_from_sha/valid_to_shaon every node and edge) plus an opt-in security scanner. The schema migration is auto-applied on firstGraphStoreopen, 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

Use thehelptool from any MCP client for inline per-tool reference.

This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.

Code Rag with Graph - local only installation

A blazingly fast codebase graphRAG implementation in 100% Rust

Recon indexes your codebase into a knowledge graph and exposes it via 14 MCP tools. AI agents get dependency mapping, blast radius analysis, safe multi-file rename, execution flow tracing, Cypher queries, semantic search, and PR review — without reading every file. Supports 13 languages, live re-index in ~50ms, and zero config setup.

Local-first knowledge graph for developers. Watches project files, extracts entities and relationships via LLMs, and lets you query across projects with natural language and source citations.

A platform-agnostic code analysis library with semantic search capabilities and MCP server support.

A server for CodeFuse-CGM, a graph-integrated large language model designed for repository-level software engineering tasks.

Generates and queries a graph representation of a codebase.

An MCP server that indexes local code into a graph database to provide context to AI assistants.

Graph-powered code intelligence MCP server with semantic search, knowledge graph, and dependency analysis for Claude Code, Cursor, and Copilot.