Graph Indexer for Cursor is a local MCP server that gives Cursor’s agent exact references, call graphs and change impact for your project. Run npx graph-indexer init and it writes .cursor/mcp.json with a graph-indexer server, then adds a short instructions block to AGENTS.md.
How do I add Graph Indexer to Cursor?
From the repository root, with Node.js 22.5 or later:
npx graph-indexer initinit configures Cursor when it finds a .cursor directory or a .cursorrules file. If Cursor is not detected, force it with npx graph-indexer init --agents cursor, or configure every supported agent at once with --all.
What does npx graph-indexer init change for Cursor?
For Cursor, init touches two files plus .gitignore:
| File | Content |
|---|---|
.cursor/mcp.json | A graph-indexer entry under mcpServers, merged with existing servers |
AGENTS.md | A managed block with the lookup rules and when to call find_references, call_graph, change_impact and check_changes |
.gitignore | .graph-indexer/, where the local index lives |
Use --dry-run to preview the changes and --no-instructions to skip the AGENTS.md block.
What does the manual Cursor configuration look like?
The entry init writes to .cursor/mcp.json is:
{
"mcpServers": {
"graph-indexer": {
"command": "npx",
"args": ["-y", "graph-indexer@3", "serve", "--repo", "${workspaceFolder}"]
}
}
}If you run Graph Indexer from a local checkout, init --local writes the absolute paths of Node.js and of that checkout’s bin/graph-indexer.mjs instead of npx.
Does Cursor run the Graph Indexer hooks?
Partly, and only if you opt in. npx graph-indexer init --hooks writes the hooks to .claude/settings.json, and the Graph Indexer README notes that Cursor also runs those Claude Code hooks. The hook command answers Cursor’s camelCase events with Cursor’s own additional_context field. Hooks are optional: the MCP tools work without them.
What does Cursor gain from Graph Indexer?
Cursor’s agent gets eight read-only tools over stdio. The ones that change its behaviour most:
find_referencesreturns every use of a symbol, bound through scopes, imports and receiver types, grouped by file and enclosing function. It says whether the list is complete.call_graphshows callers and callees across files, with confidence per edge.change_impactreturns dependents by distance and the tests to run before you change a signature.check_changesflags syntax errors, calls that no longer fit and removed names still in use, without a build.
Against the TypeScript compiler on 400 nestjs symbols, find_references reached 99.7% precision and 96.2% recall, where grep reached 12.8% precision. Graph Indexer is not a type checker: generics, reflection, any and dependency-injection tokens are only partly visible. See the tools reference and how it works.
Frequently asked questions
Why does the Cursor entry use ${workspaceFolder}?
Passing --repo ${workspaceFolder} points Graph Indexer at the workspace open in Cursor, so the server indexes that project and not whichever directory the editor happens to start it from.
Where do the lookup instructions go for Cursor?
init adds its managed block to AGENTS.md, which Cursor reads, not to a Cursor rules file. The block sits between graph-indexer:start and graph-indexer:end markers, so re-running init updates it in place.
Does Graph Indexer replace Cursor's own codebase search?
No. It adds exact tools alongside it. Cursor's search is for finding where something lives; find_references and call_graph answer which call sites are real, which a name search cannot tell apart.
Can I use it in a monorepo with Cursor?
Yes. The Twenty CRM monorepo with 28,512 files was indexed in 51 seconds at 99.9% precision and 92.4% recall for references. Later syncs of unchanged code took 1.8 seconds.