Quick start

From the root of your project, the fastest path is the one-command setup:

# init → scan → summarize, then print the Claude Code wiring line
$ orstrum quickstart

Pass -y to accept all defaults non-interactively, or --no-summarize to stop after the (free) scan step and skip the paid summarize pass.

Or run the steps individually

# 1. Write orstrum.config.ts and a CLAUDE.md guidance block
$ orstrum init

# 2. Build the graph → .orstrum/graph.db
$ orstrum scan
  scanned   : 847
  upserted  : 835
  unchanged : 12
  graph     : .orstrum/graph.db

# 3. (optional) Generate purpose summaries
$ orstrum summarize

# 4. Start the MCP server (stdio)
$ orstrum serve

Run orstrum init (or orstrum quickstart) before orstrum scan. Running scan with no orstrum.config.* present exits with "No orstrum.config.{ts,js,json} found. Run orstrum init first."
Useful init flags: -y (accept defaults), -c <path> (config location), --no-claude (skip the CLAUDE.md guidance block).

Connect to Claude Code

Add the server to your project's .claude/mcp.json:


  "mcpServers": 
    "orstrum": 
      "command": "orstrum",
      "args": ["serve"]

Claude Code launches the server over stdio automatically and can now call all eight Orstrum tools.

Verify it works

Ask Claude Code:

Who imports src/api/auth.ts?

Claude will call orstrum_get_callers and return a structured list of every file that imports that path — no grep, no file traversal.

Run orstrum scan again after any significant code change to keep the graph current. Incremental re-scans are fast — only changed files are re-parsed.