Code Review Graph MCP Server: Setup for Claude Code, Cursor and Other Apps

• Gives a tiny summary to start with

  Returns graph stats, a risk score, top areas and suggested next tools in about 100 tokens. The tool says to call it first.

  Try: “Give me a quick overview of this codebase before we start.”

  Read · get_minimal_context_tool

• Scores your changes by risk for review

  Maps your git changes to the functions, flows and areas they touch and to gaps in tests, then returns a priority-ordered review list.

  Try: “Review my uncommitted changes and tell me what to look at first.”

  Files · detect_changes_tool

• Builds a focused review pack for a change

  Combines impact analysis with source snippets and review guidance, so a review reads far less code.

  Try: “Prepare review context for my current branch.”

  Files · get_review_context_tool

• Shows what a change could break

  Lists the functions, classes and files affected by changed files. It finds changed files from git if you do not name them.

  Try: “What could break if I change auth.py?”

  Files · get_impact_radius_tool

• Answers questions like "who calls this?"

  Runs a set of ready-made queries on the graph, such as callers of a function or what references it.

  Try: “Which functions call parse_config?”

  Read · query_graph_tool

• Finds code by name, keyword or meaning

  Searches functions and classes by name or keyword, and by meaning once embeddings are built.

  Try: “Find the code that handles retry logic.”

  Read · semantic_search_nodes_tool

• Explores connections outward from a match

  Finds the best match for your question, then follows connections out to a depth you set, within a token budget.

  Try: “Trace how checkout connects to payments.”

  Read · traverse_graph_tool

• Shows which execution paths a change touches

  Finds the call chains that pass through the changed files, so you see which user-facing paths are affected.

  Try: “Which flows does my change affect?”

  Files · get_affected_flows_tool

• Lists the main execution paths

  Lists call chains that start at entry points such as web handlers and commands, most critical first.

  Try: “List the most critical flows in this app.”

  Read · list_flows_tool

• Shows one execution path in detail

  Returns the full call path of one flow with function names, files and line numbers, and optionally the code.

  Try: “Show the login flow step by step.”

  Read · get_flow_tool

• Lists clusters of related code

  Lists groups of related code found by clustering, to understand the structure at a high level.

  Try: “What are the main areas of this codebase?”

  Read · list_communities_tool

• Shows one cluster of related code

  Returns size, cohesion, main language and members of one cluster, with optional detail per member.

  Try: “Tell me about the billing cluster.”

  Read · get_community_tool

• Draws the architecture from the graph

  Builds a high-level map from the clusters, with warnings where areas depend heavily on each other.

  Try: “Give me an architecture overview with coupling warnings.”

  Read · get_architecture_overview_tool

• Finds the most connected code

  Lists the functions and classes with the most connections, where changes have the largest blast radius.

  Try: “Which parts of the code are most central?”

  Read · get_hub_nodes_tool

• Finds chokepoints between code areas

  Finds code that sits on many shortest paths between others, which would split the graph if it broke.

  Try: “Where are the architectural chokepoints?”

  Read · get_bridge_nodes_tool

• Finds weak spots in the codebase

  Lists isolated code, tiny clusters and busy code with no tests.

  Try: “Where are the untested hotspots?”

  Read · get_knowledge_gaps_tool

• Finds unexpected coupling

  Scores connections that cross areas or languages or link edge code to central code, to spot odd dependencies.

  Try: “Show me surprising dependencies between modules.”

  Read · get_surprising_connections_tool

• Suggests review questions

  Creates questions from the graph, such as which central code needs tests.

  Try: “What should I ask in this review?”

  Read · get_suggested_questions_tool

• Finds functions and files over a size

  Lists functions, classes or files above a line count, largest first.

  Try: “Find functions longer than 200 lines.”

  Read · find_large_functions_tool

• Previews renames and finds dead code

  Previews a rename across the code, finds unused code and suggests refactors. It changes nothing by itself.

  Try: “Preview renaming get_user to fetch_user.”

  Read · refactor_tool

• Applies a previewed rename to your files

  Applies the exact text replacements from an earlier preview to your source files, only inside the repo. A dry run shows a diff first, and previews expire after 10 minutes.

  Try: “Apply that rename, but show me the diff first.”

  Write · apply_refactor_tool

• Builds or refreshes the code graph

  Parses the code into the graph, updating only changed files unless you ask for a full rebuild. The graph is a SQLite file in your project.

  Try: “Build the code graph for this repo.”

  Write · build_or_update_graph_tool

• Writes a Markdown wiki of the codebase

  Creates a page for each cluster and an index in a .code-review-graph/wiki folder in your repo. It only rewrites pages that changed.

  Try: “Generate a wiki for this repo.”

  Write · generate_wiki_tool

• Reads one wiki page

  Returns the Markdown of a generated wiki page by cluster name.

  Try: “Show the wiki page for the billing cluster.”

  Read · get_wiki_page_tool

• Lists the repos you registered

  Lists repositories in the multi-repo registry stored in your home folder.

  Try: “Which repos are registered?”

  Read · list_repos_tool

• Searches across your registered repos

  Runs the search on each registered repo's graph and merges the results by rank.

  Try: “Search all my repos for uses of the Stripe client.”

  Read · cross_repo_search_tool

• Used by the assistant behind the scenes

  Recomputes flows, clusters and the search index

  Re-runs the slower analysis steps on an existing graph, such as flows, clusters and the keyword index.

  Try: “Re-run the graph analysis steps.”

  Write · run_postprocess_tool

• Builds meaning-based search for the graph

  Computes vector embeddings for every graph node so search can match by meaning. The default model runs locally; Google, OpenAI-compatible, MiniMax and Voyage providers send text to their services.

  Try: “Enable semantic search for this repo.”

  Write · embed_graph_tool

• Shows how big the graph is

  Returns counts of nodes, edges, languages and files and when the graph was last updated.

  Try: “Is the graph built and up to date?”

  Read · list_graph_stats_tool

• Reads the tool's own usage notes

  Returns one section of the bundled usage reference, such as commands or troubleshooting.

  Try: “How do I use the review commands?”

  Read · get_docs_section_tool

Watch out for

• One tool edits your source files

  apply_refactor_tool applies a previewed rename inside the repo. Use its dry run first and keep your work in git. github.com

• It writes files into your project

  The graph database and wiki live in a .code-review-graph folder in the repo, and git hooks can be installed. Add the folder to .gitignore. github.com

• Cloud embeddings send code text out

  Only if you choose a cloud provider. The default local model keeps everything on your machine. github.com

• The install step edits other apps' configs

  Its own install command writes MCP entries, hooks, skills and rule text into every supported AI tool it finds, and an uninstall command removes them. github.com

Security advisories