Live Benchmark

We compiled the Lito documentation site into a graph and ran 5 real MCP queries from Claude Code. The source docs had 6 pages annotated with graph frontmatter (type: concept and type: workflow) out of 50 total pages. Everything else was auto-detected.

Note

These are real queries and real responses — not mocked. The MCP server was running live against the compiled graph.json.

Graph Stats

Before annotations (plain Lito docs):

Metric	Value
Nodes	50
Edges	28
Node types	`doc`, `api`
Edge types	`PARENT_OF`, `CHILD_OF`

After adding 6 frontmatter annotations:

Metric	Value
Nodes	50
Edges	38
Node types	`doc`, `api`, `concept`, `workflow`
Edge types	`PARENT_OF`, `CHILD_OF`, `RELATED_TO`, `ACTS_ON`

Six annotations added 10 new semantic edges and unlocked typed navigation.

Test 1: Discovery

Query: “What concepts and APIs exist in Lito?”

Tool: list_nodes with type: "concept" and type: "api"

[
  {
    "id": "110e4edae61c",
    "type": "concept",
    "title": "Project Structure",
    "summary": "Understand the Lito project structure including special folders for assets, images, CSS, and how docs-config.json controls your site",
    "slug": "/getting-started/project-structure"
  },
  {
    "id": "2b5191a5172b",
    "type": "concept",
    "title": "Architecture",
    "summary": "Learn how the Lito CLI orchestrates template fetching, content injection, and Astro builds to generate static documentation sites",
    "slug": "/introduction/architecture"
  },
  {
    "id": "a0d625c8195b",
    "type": "concept",
    "title": "CLI Commands",
    "summary": "Complete reference for all Lito CLI commands including dev, build, init, eject, validate, doctor, info, preview, and upgrade",
    "slug": "/reference/cli-commands"
  },
  {
    "id": "49194b3d2135",
    "type": "concept",
    "title": "Configuration",
    "summary": "Full reference for docs-config.json including metadata, branding, navigation, SEO, i18n, landing pages, footer, versioning, analytics, RSS, and integrations",
    "slug": "/reference/configuration"
  }
]

[
  {
    "id": "3cf1325db142",
    "type": "api",
    "title": "Delete User",
    "summary": "Permanently delete a user account and all associated data...",
    "slug": "/api/delete-users-id"
  },
  {
    "id": "f72a8f048f74",
    "type": "api",
    "title": "Get Post Comments",
    "summary": "Retrieve all comments for a specific blog post by post ID...",
    "slug": "/api/get-posts-id-comments"
  },
  {
    "id": "48d1e17144d6",
    "type": "api",
    "title": "Get Post by ID",
    "summary": "Retrieve a single blog post by its unique identifier...",
    "slug": "/api/get-posts-id"
  },
  {
    "id": "5f392dd57765",
    "type": "api",
    "title": "List Posts",
    "summary": "Retrieve a paginated list of all blog posts...",
    "slug": "/api/get-posts"
  },
  {
    "id": "a26c4cce613e",
    "type": "api",
    "title": "Get User by ID",
    "summary": "Retrieve detailed information about a specific user...",
    "slug": "/api/get-users-id"
  },
  {
    "id": "d42a76bf78d0",
    "type": "api",
    "title": "List Users",
    "summary": "Retrieve a paginated list of all users in the system...",
    "slug": "/api/get-users"
  },
  {
    "id": "f9a2314ae2b0",
    "type": "api",
    "title": "Create Post",
    "summary": "Create a new blog post with title, body, and author details...",
    "slug": "/api/post-posts"
  },
  {
    "id": "959f5793dcf9",
    "type": "api",
    "title": "Create User",
    "summary": "Create a new user account with email, password, and optional profile...",
    "slug": "/api/post-users"
  },
  {
    "id": "6f61d691b413",
    "type": "api",
    "title": "Update User",
    "summary": "Update an existing user's profile information, email, role...",
    "slug": "/api/put-users-id"
  }
]

Verdict: Agent instantly gets a typed inventory — 4 concepts, 9 APIs — with summaries and navigable slugs. No text chunking, no guessing.

Test 2: Relationship Traversal

Query: “What’s related to the Configuration concept?”

Tool: traverse with node_id: "49194b3d2135" (Configuration)

[
  {
    "id": "110e4edae61c",
    "type": "concept",
    "title": "Project Structure",
    "edge_types": ["RELATED_TO"]
  },
  {
    "id": "e41b56795af6",
    "type": "workflow",
    "title": "Quick Start",
    "edge_types": ["ACTS_ON"]
  },
  {
    "id": "112cbdc50459",
    "type": "workflow",
    "title": "Adding Pages",
    "edge_types": ["ACTS_ON"]
  },
  {
    "id": "2b5191a5172b",
    "type": "concept",
    "title": "Architecture",
    "edge_types": ["RELATED_TO"]
  },
  {
    "id": "a0d625c8195b",
    "type": "concept",
    "title": "CLI Commands",
    "edge_types": ["RELATED_TO"]
  }
]

Verdict: From a single node, the agent discovers 5 connected nodes across 2 edge types. It knows that the Quick Start and Adding Pages workflows both act on Configuration, and that Architecture, CLI Commands, and Project Structure are related concepts.

Tip

This is impossible with RAG. A vector search for “Configuration” would return text chunks — not typed relationships to workflows and sibling concepts.

Test 3: Workflow Planning

Query: “How do I set up a new Lito site?”

Tool: list_nodes with type: "workflow", then get_workflow on Quick Start

[
  {
    "id": "e41b56795af6",
    "type": "workflow",
    "title": "Quick Start",
    "summary": "Create your first Lito documentation site in minutes...",
    "slug": "/getting-started/quick-start"
  },
  {
    "id": "112cbdc50459",
    "type": "workflow",
    "title": "Adding Pages",
    "summary": "Add and organize documentation pages in Lito...",
    "slug": "/guides/adding-pages"
  }
]

{
  "workflow": {
    "title": "Quick Start",
    "goal": "Set up a new Lito documentation site from scratch.",
    "primary_entity": "Configuration",
    "risk_level": "low",
    "requires_human_approval": false,
    "url": "https://lito.rohitk06.in/getting-started/quick-start",
    "anchors": [
      "quick-start",
      "1-prepare-your-content",
      "hello-world",
      "2-add-configuration",
      "3-run-the-dev-server",
      "4-view-your-site",
      "whats-next"
    ]
  }
}

Verdict: The agent gets structured workflow metadata — goal, risk level, approval requirements, and section anchors for step-by-step navigation. It can link the user directly to https://lito.rohitk06.in/getting-started/quick-start#2-add-configuration.

Test 4: Semantic Search

Query: “What references Configuration?”

Tool: search with query: "Configuration"

[
  {
    "id": "49194b3d2135",
    "type": "concept",
    "title": "Configuration",
    "summary": "Full reference for docs-config.json including metadata, branding, navigation, SEO, i18n...",
    "score": 23
  },
  {
    "id": "d65bee9f1c74",
    "type": "doc",
    "title": "FAQ",
    "summary": "Frequently asked questions about Lito — installation, configuration, framework support...",
    "score": 5
  },
  {
    "id": "112cbdc50459",
    "type": "workflow",
    "title": "Adding Pages",
    "summary": "Add and organize documentation pages in Lito with file-system routing, frontmatter metadata...",
    "score": 5
  },
  {
    "id": "110e4edae61c",
    "type": "concept",
    "title": "Project Structure",
    "summary": "Understand the Lito project structure including special folders for assets, images, CSS...",
    "score": 3
  }
]

Verdict: Ranked results across node types. The agent knows that changing Configuration impacts the Adding Pages workflow and the Project Structure concept — instant impact analysis.

Test 5: Deep Node Intelligence

Query: “Tell me everything about the CLI.”

Tool: get_node with identifier: "a0d625c8195b"

{
  "title": "CLI Commands",
  "type": "concept",
  "entity_type": "tool",
  "canonical_name": "CLI",
  "aliases": ["Lito CLI", "lito command", "@litodocs/cli"],
  "related_entities": ["Configuration", "Template"],
  "url": "https://lito.rohitk06.in/reference/cli-commands",
  "anchors": [
    "cli-commands", "init", "interactive-mode", "with-options", "options",
    "build", "with-global-installation", "with-npx", "dev", "eject",
    "preview", "preview-existing-build", "build-and-preview",
    "validate", "doctor", "info", "template-list", "template-cache",
    "upgrade"
  ],
  "tags": ["reference", "cli"]
}

Verdict: The agent gets:

Canonical name + aliases — recognizes “Lito CLI”, “lito command”, and “@litodocs/cli” as the same entity
Entity type — knows it’s a tool, not a concept or resource
50+ section anchors — can link to exact sections like #init, #build, #validate
Public URL — https://lito.rohitk06.in/reference/cli-commands
Related entities — Configuration and Template

Results Summary

Test	Query	What the agent gets
Discovery	List concepts + APIs	Typed inventory with summaries and slugs
Traversal	What’s related to Configuration?	5 connected nodes via `RELATED_TO` + `ACTS_ON`
Workflow	How to set up a Lito site?	Goal, risk level, approval flag, step anchors, URL
Search	What references Configuration?	Ranked hits across types with relevance scores
Deep Intel	Everything about the CLI	Aliases, entity type, 50+ anchors, public URL

Note

All 5 tests ran against the real Lito documentation at lito.rohitk06.in — not synthetic test data. Only 6 out of 50 pages were annotated with graph frontmatter. The 9 API nodes were auto-detected from existing api: "GET /endpoint" format.

Key Takeaway

Six frontmatter annotations turned a flat file tree into a navigable knowledge graph with typed relationships. An agent using this graph doesn’t search — it navigates. It doesn’t guess — it traverses. And every node links to a public URL the user can visit.

5 min read