Open Source · MIT License

Knowledge graphs for AI agents

Turn your Markdown docs into a typed knowledge graph with nodes, edges, and workflows. Query it via MCP so AI agents actually understand your documentation.

USES_APIRELATED_TOACTS_ONNEXT_STEP_OFDoctype: docAPItype: apiConcepttype: conceptWorkflowtype: workflow

Three steps to agent-native docs

Write Markdown

Add typed frontmatter to your existing docs. Five node types, simple YAML.

Build Graph

The CLI compiles your docs into a typed JSON graph with nodes and edges.

Query via MCP

AI agents search, traverse, and discover workflows through 7 MCP tools.

Structured knowledge, not flat files

doc

Doc

Guides, tutorials, reference pages

concept

Concept

Domain terms, entities, definitions

api

API

Endpoints, functions, tool specs

workflow

Workflow

Multi-step procedures with guardrails

step

Step

Individual actions in a workflow

Relationships

  • RELATED_TO
  • DEPENDS_ON
  • PART_OF
  • EXTENDS

Actions

  • ACTS_ON
  • USES_API
  • CONFIGURED_BY
  • TRIGGERED_BY
  • PRODUCES

Workflows

  • HAS_STEP
  • NEXT_STEP_OF
  • ALTERNATIVE_TO
  • REQUIRES
  • VALIDATES
list_nodes
Filter nodes by type or tag
get_node
Full details by ID or slug
traverse
Find connected nodes by edge type
search
Keyword search across all fields
get_workflow
Steps, API links, and guardrails
find_apis_for_entity
APIs that act on a concept
get_graph_stats
Overview counts and metadata

From zero to MCP in one command

Run the CLI, point it at a graph JSON, and your docs are queryable by any MCP client.

Terminal
~ |
Fetching graph...
Loaded: 50 nodes, 38 edges
MCP server running on stdio
AI Agent
// Agent calls search tool
search({ "query": "authentication" })
// Returns matching nodes
[{
  "type": "api",
  "title": "POST /auth/login",
  "tags": ["auth", "jwt"]
}]

Built for the agent era

Zero Install

Run with npx. No global packages, no setup wizard, no config files to maintain.

Any Docs Framework

Works with any Markdown-based docs. Astro, Next.js, Docusaurus, VitePress, or plain files.

Graph Traversal

Agents navigate typed edges to discover related concepts, APIs, and workflow steps.

Workflow Safety

Workflows include guardrails, required checks, and ordered steps agents must follow.

Public Hosting

Host your graph.json on any CDN or static host. Agents fetch it at runtime via URL.

Fully Local

Everything runs on your machine. No cloud service, no API keys, no data leaves your network.

5 Node Types
14 Edge Types
7 MCP Tools
MIT Licensed

Ready to graph your docs?

Give your AI agents structured knowledge instead of raw text. Get started in one command.

$ npx @litodocs/graph https://your-docs.com/graph.json