DEV Community

Graphify Labs Inc profile image Syed Fahad
Syed Fahad for Graphify Labs Inc

Posted on

Give Your AI Coding Agent a Map of Your Codebase (with Graphify)

#ai #productivity #tutorial #opensource

TL;DR: Graphify is an open source CLI that turns your codebase into a knowledge graph on your own machine. Install it with uv, pipx or pip, map your repo, and your coding agent can follow how your code actually connects instead of grepping in circles.

AI coding agents are great at writing code. Where they struggle is finding their way around a real codebase. Ask about a feature on a big repo and you can watch it happen: grep, open a file, grep again, reread something it already read.

The missing piece is a map. A knowledge graph turns files, functions and concepts into nodes, and the relationships between them (calls, imports, depends on) into edges. Your agent can then jump to the right place and follow the connections.

Here's how to set one up with Graphify.

1. Install

Use whichever installer you like. You need Python 3.10 or later.

# uv
uv tool install graphifyy

# pipx
pipx install graphifyy

# pip, inside a virtual environment
python3 -m venv .venv
source .venv/bin/activate      # Windows: .venv\Scripts\activate
python -m pip install graphifyy
Enter fullscreen mode Exit fullscreen mode

The package is graphifyy (two y's) and the command is graphify. Check it worked:

graphify --version
Enter fullscreen mode Exit fullscreen mode

2. Connect it to your assistant 

graphify install
Enter fullscreen mode Exit fullscreen mode

That registers the Graphify skill with the assistant it detects. To choose one explicitly:

graphify install --platform cursor   # or claude, codex, gemini, copilot
Enter fullscreen mode Exit fullscreen mode

For Copilot in VS Code, use graphify vscode install. Copilot CLI uses graphify install --platform copilot.

3. Map your repo

In your assistant's chat, from the root of the repo:

/graphify .
Enter fullscreen mode Exit fullscreen mode

In Codex, select the skill with $graphify and ask it to map the current repo.

You can also skip the assistant and build a code graph from the terminal. Code only extraction doesn't need an API key:

graphify extract . --code-only
Enter fullscreen mode Exit fullscreen mode

Either way you end up with a graphify-out/ folder:

graphify-out/
├── graph.html        Interactive graph, open it in a browser
├── GRAPH_REPORT.md   Main concepts, communities, suggested questions
└── graph.json        The graph the CLI and MCP tools query
Enter fullscreen mode Exit fullscreen mode

Open graph.html and click around. Then skim GRAPH_REPORT.md, it's a surprisingly good tour of your own project.

4. Ask it things

From the terminal:

graphify query "how does a request reach the database?"
Enter fullscreen mode Exit fullscreen mode

The result is graph context: nodes, the relations between them, and the file references to open. It is not a verified answer by itself. The default budget is 2000 tokens. Narrow a long result with --budget 1500, or ask about something smaller.

Or just ask your agent in chat: "using the graph, what depends on the auth module?"

One honest note: this is context from the files Graphify analyzed. Open the lines it points to before you rely on it. Every edge is tagged EXTRACTED, INFERRED or AMBIGUOUS, so you can tell what was parsed from what was guessed.

5. Optional: connect over MCP

If you'd rather your agent call the graph tools directly, use the local MCP server. Install the MCP extra (keep the quotes, zsh needs them):

uv tool install "graphifyy[mcp]"
Enter fullscreen mode Exit fullscreen mode

Already installed without it? Run uv tool install --reinstall "graphifyy[mcp]".

Then register it, always with an absolute path to the graph.

Claude Code:

claude mcp add --transport stdio graphify -- graphify-mcp /absolute/path/to/repo/graphify-out/graph.json
Enter fullscreen mode Exit fullscreen mode

Codex:

codex mcp add graphify -- graphify-mcp /absolute/path/to/repo/graphify-out/graph.json
Enter fullscreen mode Exit fullscreen mode

Cursor, in your user or project MCP config:

{
  "mcpServers": {
    "graphify": {
      "command": "graphify-mcp",
      "args": ["/absolute/path/to/repo/graphify-out/graph.json"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Refresh your MCP connections and ask the agent to call graph_stats, then query_graph with a question about your repo. Other clients are in the MCP docs.

6. Keep the graph fresh 

graphify hook install
Enter fullscreen mode Exit fullscreen mode

Install that once in each clone. After you pull or merge, refresh the map:

graphify update .
Enter fullscreen mode Exit fullscreen mode

Troubleshooting

  • graphify not found: run uv tool update-shell or pipx ensurepath, then open a new terminal.
  • An MCP client can't find graphify-mcp: use the absolute path to the executable in its config.
  • zsh complains about brackets: wrap extras in quotes, like "graphifyy[mcp]".

Try this

Ask your agent the same question with and without the graph, something like "where is user input validated before it hits the database?" Notice how many files it opens each time. That comparison tells you more than any description.

Full docs are at docs.graphify.com. Graphify is open source, so if it helps you, a star on GitHub helps other developers find it.

Top comments (0)