Quick Start

Build a graph from your docs and start querying it in under 2 minutes.

Option A: Use a hosted graph (zero-install)

If your docs site already serves graph.json, connect any MCP client with one command — no cloning, no dependencies:

Terminal window
npx -y @litodocs/graph https://docs.example.com/graph.json

Skip to step 5 to wire it into your IDE.

Option B: Build locally

1. Install

Terminal window
cd lito-graph
bun install

2. Build a Graph

Point it at any folder of Markdown files:

Terminal window
lito-graph build -i ./path/to/docs -o ./graph.json

You’ll see output like:

◇ Found 12 nodes, 15 edges
◇ Graph written to graph.json
Node breakdown:
api: 3
concept: 2
doc: 2
workflow: 1
step: 4

3. Inspect the Graph

Terminal window
# Overview stats
lito-graph inspect -g ./graph.json --stats
# List all nodes
lito-graph inspect -g ./graph.json --nodes
# Filter by type
lito-graph inspect -g ./graph.json --type api
# Find broken references
lito-graph inspect -g ./graph.json --unresolved

4. Start the MCP Server

Terminal window
lito-graph serve -g ./graph.json

Or build and serve in one step:

Terminal window
lito-graph serve -i ./path/to/docs

5. Connect to Claude Code

Create .mcp.json in your project root:

{
"mcpServers": {
"lito-graph": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@litodocs/graph", "https://docs.example.com/graph.json"]
}
}
}

Start a new Claude Code session. You’ll have access to all 7 graph tools.

Next Steps

  • Writing Graph Docs — add type: concept, type: api, and type: workflow to your frontmatter
  • MCP Tools — explore the tools agents can use