Agent Discovery

This workflow describes the recommended pattern for AI agents to use the Lito Graph MCP tools.

Preconditions

  • Lito Graph MCP server is running and connected to the agent’s IDE.
  • A graph has been built from the project’s documentation.

Steps

  1. Call get_graph_stats to understand the scope of the documentation.
  2. Call list_nodes with type filters to discover available concepts, APIs, and workflows.
  3. Call search with the user’s intent to find relevant nodes.
  4. Call get_node to read full details of matching nodes.
  5. Call traverse to explore connected nodes and understand relationships.
  6. Call get_workflow if a multi-step procedure is needed.
  7. Call find_apis_for_entity to discover all capabilities for a given entity.

Failure Modes

  • Search returns no results because docs lack graph frontmatter annotations.
  • Workflow steps reference APIs that don’t have corresponding API docs.

Recovery

  • If search returns empty: fall back to listing all nodes and filtering manually.
  • If unresolved API references exist: warn the user that documentation is incomplete.

Guardrails

  • Always check requires_human_approval on workflows before executing steps.
  • Always check risk_level and present it to the user for medium or high risk workflows.
  • Never execute API calls without first reading their preconditions and permissions.
  • Present guardrails from workflow docs to the user before proceeding.
1 min read