Usage

◈ Claude Desktop

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "code-crawler": {
      "command": "node",
      "args": [
        "/absolute/path/to/code-crawler/dist/index.js"
      ],
      "env": {
        "CODECRAWLER_DB": "/absolute/path/index.db"
      }
    }
  }
}

macOS: ~/Library/Application Support/Claude/ · Windows: %APPDATA%\Claude\ · Linux: ~/.config/Claude/

◈ Claude Code

One command registers the server:

claude mcp add code-crawler -- \
  node /absolute/path/to/code-crawler/dist/index.js

Verify with claude mcp list, then run /mcp inside a session.

◈ Cursor

Project .cursor/mcp.json or global ~/.cursor/mcp.json:

{
  "mcpServers": {
    "code-crawler": {
      "command": "node",
      "args": [
        "/absolute/path/to/code-crawler/dist/index.js"
      ]
    }
  }
}

◈ VS Code / Copilot

Workspace .vscode/mcp.json uses a servers map:

{
  "servers": {
    "code-crawler": {
      "type": "stdio",
      "command": "node",
      "args": [
        "/absolute/path/to/code-crawler/dist/index.js"
      ]
    }
  }
}

◆ GitHub Copilot app

Open Customize → MCP. If Code-Crawler appears in the catalog, select it and follow the prompts; otherwise add a custom server named code-crawler with this command:

npx -y @nithin554/code-crawler@latest

The app stores the entry in ~/.copilot/mcp-config.json and lists it under Customize → Installed. Sessions already running keep their current tool set — the server is spawned for the next session or after you toggle it off and on.

◆ Copilot CLI

One command writes the user-level configuration:

copilot mcp add code-crawler -- \
  npx -y @nithin554/code-crawler@latest

Verify with copilot mcp list, or /mcp inside a session. Add env vars with --env CODECRAWLER_DB=....

◆ VS Code

If the catalog lists it, install from the Extensions view with @mcp code-crawler. Otherwise add the entry to .vscode/mcp.json (workspace) or your user MCP settings:

{
  "servers": {
    "code-crawler": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@nithin554/code-crawler@latest"]
    }
  }
}

◆ Other stdio clients

Any client that accepts a command and args works — no local build and no absolute path to dist/index.js:

{
  "mcpServers": {
    "code-crawler": {
      "command": "npx",
      "args": ["-y", "@nithin554/code-crawler@latest"],
      "env": {
        "CODECRAWLER_DB": "/absolute/path/index.db"
      }
    }
  }
}

1. Index explicitly

Start a session by asking the assistant to crawl the project at an absolute path. It returns file, node, edge, and call-resolution counts you can sanity check.

crawl_project({ path: "C:/work/shop-service" })

2. Resolve names first

Tool arguments take symbol ids, not guesses. Search once, then reuse the returned qualifiedName.

search_symbols({ query: "place" })

3. Ask one graph question at a time

Prefer a specific tool over a vague prompt. “Who calls X?” maps to get_callers; “what runs?” maps to trace_execution.

get_callers({
  symbol: "com.acme.OrderService#place",
  transitive: true
})

4. Ask for file and line proof

Results carry file and callSiteLine (1-based in output). Ask the assistant to cite them before editing.