Verify setup

Check the CLI

larkx --version
# → larkx v0.1.0

If the command is not found, make sure your Node global bin is on PATH.

Check the index

larkx stats

Should print file counts, detected frameworks, and per-level token estimates.

If the index is empty, run larkx index first.

Check the MCP server

larkx mcp --check
# → ✓ MCP server responding correctly

This confirms the MCP server can start and answer basic JSON-RPC health checks.

Check the context file

cat .larkx/context.md

The file should exist and contain a compact project snapshot. If it is missing or empty, run larkx index.

Check inside your AI agent

Open VS Code with this project. Ask Claude Code:

"Which MCP servers do you have?"

You should see larkx in the list with its 6 tools.

Run a tool diagnostic

Ask the AI to use larkx tools instead of opening source files:

1. Call get_project_index and report the file count.
2. Pick a function name from that result and call search_symbol.
3. Call get_file_summary on any indexed file.
4. Call get_dead_code and report how many dead nodes are found.

The agent should return real data for every step and should not read raw source files first.