graphify: Turn a Codebase into a Knowledge Graph for Claude
Maps your code, docs and papers into a knowledge graph your assistant can query, so it answers architecture questions without reading every file.
- Installs
- 12.1K
- on skills.sh
- Stars
- 140.5K
- on GitHub
- Files
- 60+
- besides SKILL.md
- Updated
- Nov 2026
- last repo commit
graphify makes your coding assistant map a project into a knowledge graph, then answer questions from that graph instead of searching file by file. You run the /graphify command on a folder and it builds a clickable graph and a short report of the main ideas in your project. Code is read on your machine, but docs, PDFs and images are sent to your assistant's model. The skill is only the front door to a Python tool, so the install box above installs the graphifyy package first.
Good for
- Developers who join big codebases and want a map before they read any code
- Claude Code, Codex or Cursor users who want the assistant to check a graph before it searches files
- Researchers who keep papers, notes and screenshots in one folder and want to see how they connect
- Teams that review many pull requests and want to see what each change touches
Install graphify
Claude Code
- 1
Copy this prompt.
Install the graphify skill in Claude Code. Run this command: uv tool install graphifyy graphify install Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Python 3.10 or newer, uv (or pipx).
The PyPI package is graphifyy, with two y's; the command it installs is graphify. graphify install registers the skill and writes Claude instructions and PreToolUse hooks that point Claude Code at the graph. Then open your project and type /graphify . to build the graph. Update with uv tool upgrade graphifyy, then run graphify install --platform claude again.
From the project README. Source.
What's in the skill
When it loads
Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community detection, and query/path/explain tools.
Sections in SKILL.md
- Usage
- What graphify is for
- What You Must Do When Invoked
- Step 0 - GitHub repos and multi-path merge (only if a URL or several paths)
- Step 1 - Ensure graphify is installed
- Step 2 - Detect files
- Step 2.5 - Video and audio (only if video files detected)
- Step 3 - Extract entities and relationships
- Step 4 - Build graph, cluster, analyze, generate outputs
- Step 4.5 - Graph health check (read-only integrity gate)
- Step 5 - Label communities
- Step 6 - Generate Obsidian vault (opt-in) + HTML
- Steps 6b-8 - Wiki, Neo4j, FalkorDB, SVG, GraphML, MCP, benchmark (only on their flags)
- Step 9 - Save manifest, update cost tracker, clean up, and report
- Interpreter guard for subcommands
- For --update and --cluster-only
Key facts
- License
- Apache-2.0
- Installs on skills.sh
- 12,102 (Nov 1, 2026)
- GitHub stars
- 140,465 (Nov 1, 2026)
- What installs
- The skills CLI copies the skill file only. The tool itself is the PyPI package graphifyy.
- Works in
- Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, VS Code Copilot Chat, OpenCode and more
- skills.sh audits
- Snyk pass, Gen Agent Trust Hub pass, Socket warn
- Last commit
- Nov 1, 2026
What it includes
Maps the whole folder into a knowledge graph and opens a report
Runs the full pipeline on the current folder: finds files, parses code locally, has the assistant read docs, papers and images, clusters the result into communities and writes graph.html, GRAPH_REPORT.md and graph.json into graphify-out/. It then shows the key concepts and offers a question to explore.
Try: “/graphify .”
Runs code ·
/graphify .Clones a GitHub repo (or several) and builds one graph
Give it one or more GitHub URLs, optionally with --branch, and it clones them and builds a graph; several repos are merged into one cross-repo graph.
Try: “/graphify https://github.com/<owner>/<repo>”
Runs code ·
/graphify <github-url>Answers a question from the graph instead of reading files
Searches the graph for nodes matching your question and returns the connected piece of the graph, broad (default) or as one traced path with --dfs. --budget caps the answer size in tokens. Answers cite source locations.
Try: “/graphify query "what connects auth to the database?"”
Read ·
/graphify queryShows the shortest chain of links between two things
Finds how two names connect, hop by hop, and reports the path's overall confidence (its weakest link).
Try: “/graphify path "UserService" "DatabasePool"”
Read ·
/graphify pathExplains one function, class or concept and what it connects to
Prints where a node is defined, its community, how many links it has and each connection tagged EXTRACTED or INFERRED.
Try: “/graphify explain "RateLimiter"”
Read ·
/graphify explainRe-reads only the files that changed since the last build
Incremental rebuild: re-extracts new or changed files and keeps the rest from cache. Code changes need no model calls.
Try: “/graphify ./docs --update”
Write ·
--update / graphify updateRegroups the existing graph without re-reading files
Reruns community detection on the graph you already have and rewrites the report; --resolution makes groups finer and --exclude-hubs hides utility hubs from the top concepts.
Try: “/graphify . --cluster-only --resolution 1.5”
Write ·
--cluster-onlyMore thorough extraction with more inferred links
Tells the extraction subagents to dig harder and add more INFERRED edges, at higher token cost.
Try: “/graphify <path> --mode deep”
Write ·
--mode deepFetches a web page, paper or video and adds it to the graph
Downloads the URL into ./raw (transcribing videos), then updates the graph. --author and --contributor tag who wrote or added it.
Try: “/graphify add https://arxiv.org/abs/1706.03762”
Write ·
/graphify add <url>Rebuilds the code graph every time you save a file
Watches the folder and rebuilds the code part of the graph on changes with no model calls. Needs the watch extra.
Try: “graphify watch .”
Write ·
--watch / graphify watchWrites the graph as an Obsidian vault, wiki, SVG, GraphML or database import
Optional outputs on their flags: an Obsidian vault, a Markdown wiki with one article per community, graph.svg, GraphML for Gephi or yEd, and Cypher files or a direct push to Neo4j or FalkorDB.
Try: “/graphify . --wiki”
Write ·
Exports (--obsidian, --wiki, --svg, --graphml, --neo4j, --falkordb)Draws an architecture and call-flow page
Writes a Mermaid-based HTML page of the architecture and call flow; with the git hook installed it regenerates on every commit.
Write ·
graphify export callflow-htmlServes the graph to any MCP client as 10 tools
Starts an MCP server over stdio (or HTTP with an API key) that reads graph.json and gives your agent tools like query_graph, get_node, get_neighbors, shortest_path, god_nodes and PR-impact tools. Install with graphifyy[mcp].
Try: “claude mcp add --transport stdio graphify -- graphify-mcp /absolute/path/to/repo/graphify-out/graph.json”
Network ·
--mcp / graphify-mcpMakes your assistant check the graph before it greps or reads files
Writes a graphify section into CLAUDE.md, AGENTS.md, GEMINI.md or a Cursor rule, plus hooks on Claude Code, Codex and Gemini CLI that nudge the agent toward graphify query. --strict blocks the first raw read per session.
Try: “graphify claude install”
Write ·
graphify <platform> install (always-on)Rebuilds the graph after commits, checkouts and merges
Adds git hooks for commit, checkout, merge and rebase that rebuild the code graph, and a merge driver so graph.json conflicts resolve on their own.
Try: “graphify hook install”
Runs code ·
graphify hook installBuilds the graph from a terminal or CI without an assistant
Headless build. --code-only skips docs and media so it runs fully offline; otherwise it needs a provider key (Gemini, Claude, OpenAI, Ollama and others) for the semantic pass.
Try: “graphify extract . --code-only”
Write ·
graphify extractLists what depends on a node, or the most connected nodes
affected walks links backwards to show what a change to one node reaches; god-nodes lists the most connected concepts.
Read ·
graphify affected / god-nodesPull request dashboard with graph impact
Shows CI and review state per PR, a deep dive on one PR's graph impact, --triage to rank the review queue with your configured model, and --conflicts for PRs touching the same communities.
Try: “graphify prs 42”
Runs code ·
graphify prsCombines graphs from several repos
merge-graphs joins two graph.json files; global add/list/remove manages one global graph at ~/.graphify/global-graph.json.
Try: “graphify merge-graphs a.json b.json”
Write ·
graphify merge-graphs / globalRecords what worked so later sessions learn from it
save-result stores a question, answer and outcome; reflect writes graphify-out/reflections/LESSONS.md.
Write ·
graphify save-result / reflectRemoves the skill and rules from every assistant
Removes graphify from all platforms at once; --purge also deletes graphify-out/. Per-platform versions exist, e.g. graphify claude uninstall.
Try: “graphify uninstall --purge”
Write ·
graphify uninstall
From the official docs
Permissions
Medium risk
graphify runs a Python package that reads your repo and writes hooks into your assistant's config. Code stays local; docs, PDFs and images go to your assistant's model.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only modeHow to turn it on
Watch out for
Re-running the Claude install can delete CLAUDE.md notes
An open issue reports that it removes everything below the graphify section when no later top-level heading exists. Back up CLAUDE.md first. github.com
skills.sh: Socket warn
One low anomaly alert on skill.md (Oct 16, 2026). skills.sh
skills.sh: Gen Agent Trust Hub pass
Rated safe (Oct 16, 2026). Notes subprocess and git calls, network fetches only for URLs you add, git hooks and skill files written only when you run the install commands. skills.sh
skills.sh: Snyk pass
Risk level low, no issues detected (Oct 16, 2026). skills.sh
Technical details
| Turning on read-only | The optional graphify-mcp server exposes 10 read-only tools over a graph you already built. |
|---|---|
| Talks to | Your assistant's model: docs, PDFs and images go there for the semantic pass. Code files stay on your machine.. Headless graphify extract uses whichever API key is set (Gemini, Kimi, Claude, OpenAI, DeepSeek and others). GRAPHIFY_BACKEND=none turns all model calls off.. URLs you add with /graphify add (arXiv, YouTube, X). No telemetry, per the README |
| Scopes |
|
What people say
People are curious about the idea and unconvinced it beats plain search on strong models.
What people like
- A user comparing it with Ix for their project picked graphify.
- Other tools build on its output, and Graphify C# says its JSON can be used directly by graphify.
Common complaints
- Hacker News commenters doubt code-graph tools beat a plain agent with ripgrep on current models.
- The benchmark numbers in the vendor's discussion came from a Graphify Labs collaborator, so they are not independent.
- Copilot CLI users report it picks up API keys from the environment instead of using their Copilot plan.
In their words
“I read through the skill file of graphify and ran some of the commands it was telling the agent to run. The vibe I got is there was a lot of structured information returned that the model (and harness) isn't set up to work with effectively.”NewJazzHacker NewsOct 2026
“Has anyone actually gotten good results out of this? For the evals I've run (albeit, not super scientific), these memory systems did worse than the raw harness given ripgrep.”theLiminatorGitHub issueSep 2026
“I run graphifyy inside GH Copilot in order to make it consume the GH Copilot subscription, not my direct OpenAI one.”rakotomandimby
Examples
Build a graph of the current project
/graphify .
SourceAsk how two parts of the code connect
/graphify query "what connects auth to the database?"
SourceTrace the shortest path between two classes
/graphify path "UserService" "DatabasePool"
SourceAsk Claude Code about a flow once the graph exists
Use Graphify to explain the authentication flow. Cite the source files and distinguish explicit relationships from inferred ones.
SourceAsk a plain question that the hooks route to the graph
where do we retry failed uploads?
SourceAdd a research paper to the graph
/graphify add https://arxiv.org/abs/1706.03762
Source
Questions
How do I install graphify?
Install the Python package graphifyy (double y) with uv tool install graphifyy or pipx install graphifyy, then run graphify install so it registers the skill with your assistant. Open your project and type /graphify .. It needs Python 3.10 or newer. Other assistants use their own flag, such as graphify install --platform codex; the install box has each one.
Does graphify work in Claude.ai, Codex and Cursor?
Claude Code, Codex and Cursor are documented. Codex runs it as $graphify and needs multi_agent = true in its config for parallel work, and Cursor gets a short rule file rather than the full skill. No source covers the claude.ai chat site or the Claude desktop chat, so do not count on it there.
Is graphify safe?
Mostly. On skills.sh, Snyk and Gen Agent Trust Hub pass it and Socket warns with one low finding. One advisory, GHSA-pcc4-rvhr-2pr8, is fixed in 0.9.70; the repo rates it high and the GitHub Advisory Database rates it moderate. The permissions section above lists what it reads, writes and sends.
Can graphify delete my CLAUDE.md notes?
It can. An open issue reports that re-running the Claude install deletes everything in CLAUDE.md below the graphify section when no later top-level heading exists, with no warning. Back up CLAUDE.md before you re-run the install, or put your own notes under a top-level heading after the graphify block.
Does graphify send my code anywhere?
Code files are read on your machine and nothing leaves it for those, and the README says there is no telemetry. Docs, PDFs and images go to your assistant's model, or to Gemini if a Gemini or Google key is set. Setting GRAPHIFY_BACKEND=none turns all model calls off.
Is graphify free, and does it need an API key?
Yes, it is free and open source under Apache-2.0, and the skill needs no API key. It uses your assistant's own model for docs and images, so reading many papers or images can take minutes and tens of thousands of tokens. Code-only runs cost nothing in model use. Graphify Cloud is a separate hosted product.
What is the Karpathy connection?
Graphify's first PyPI release opened with Andrej Karpathy's /raw folder of papers, tweets and notes, which gets hard to see into over time, and called graphify the answer to that problem. That is the link the sources show.
Does graphify have an MCP server?
Yes, as an optional extra. Install the package with its mcp extra, build a graph, then point your client at graphify-mcp and the path to graph.json. It gives your assistant 10 read-only tools and runs on your machine unless you start it in HTTP mode.
graphify vs Ix or Graphify Cloud?
Ix is another open-source command line tool that maps a codebase into a persistent graph for agents, and it was pitched on Hacker News as the alternative to graphify's inferred links. Graphify Cloud is the same team's hosted version, which keeps an always-on graph across repos instead of building graphify-out locally.
Does graphify actually save tokens?
The vendor reports 82% versus 70.8% key-fact coverage over a grep and read baseline on ERPNext. Hacker News commenters are skeptical and say plain ripgrep did as well in their informal tests. The skill itself says a plain grep is faster for a one-off lookup.
Information
updated Nov 2, 2026
- Repository
- graphify-labs/graphify
- Folder
graphify
Sources
Docs, issue trackers and discussions this page is built from.
Show all 25Hide
- graphify README · github.com
- Graphify for Claude Code · docs.graphify.com
- Graphify integrations · docs.graphify.com
- graphify skill.md · github.com
- Author profile · github.com
- License · github.com
- graphify on skills.sh · skills.sh
- Repository · github.com
- graphifyy on PyPI · pypi.org
- Graphify always-on setup · docs.graphify.com
- Graphify MCP server · docs.graphify.com
- docs.graphify.com · docs.graphify.com
- Socket · skills.sh
- Snyk · skills.sh
- Gen Agent Trust Hub · skills.sh
- Security advisory GHSA-pcc4-rvhr-2pr8 (repo, severity high) · github.com
- GitHub Advisory Database: GHSA-pcc4-rvhr-2pr8 (severity moderate) · github.com
- Re-running install deletes CLAUDE.md content · github.com
- graphifyy 0.1.1 · pypi.org
- Ix · github.com
- Ask HN: We got 100 stars · news.ycombinator.com
- Benchmarks discussion · github.com
- Show HN: Graphicle, a Rust port of Graphify · news.ycombinator.com
- Graphify C# · news.ycombinator.com
- Copilot CLI issue · github.com