Capability

Ground your AI agents in a structured, design-time model — not RAG over docs.

MCP servers expose blueprint, brandvoice, realm, hero, and gamemaster models as inference-time context. Structured grounding lifts accuracy 20%+ and cuts tool calls by 39%.

What it is

AI grounding means providing agents with structured, queryable, business-semantic context at inference time — not as RAG retrieval over documentation, but as explicit machine-readable model state surfaced via Model Context Protocol (MCP) tools. Five+ MCP servers ship today: blueprint MCP (entities, relations, metadata, modelDiff), brandvoice MCP (~21 domain tools), realm MCP (spatial, maintenance, compliance), hero MCP (career timeline, skill matrix), and gamemaster MCP. The grounding category is closing the context-management infrastructure gap that Gartner projects will cancel ~half of agentic AI projects by 2027.

Built on

  • Blueprint MCP — entities, relations, metadata, modelDiff
  • Brandvoice MCP — ~21 domain tools (positioning, content priorities, evidence coverage)
  • Realm MCP — spatial, maintenance calendar, compliance, migration impact
  • Hero MCP — career timeline, skill matrix, portfolio summary, evidence trail
  • gamemaster + sdlc MCP — board-game tooling + plan-query tooling

How it works

Author the model. Stand up the MCP server (blueprint MCP is the reference implementation). Point your agent runtime at the MCP endpoint. The agent now queries the model at inference time — pulls entities, relations, validated metadata, and structured context for the question at hand. No vector index, no RAG fragility, no doc drift. The model is the same one your team uses for architecture decisions and your generators use for API contracts.

Query the brandvoice MCP from the playground →

AI agents need an 'ontology' to act as a shared rulebook so they don't make confident, expensive mistakes.
— Nathan Lasnoski, “9 things for CIOs to focus on in 2026” (December 2025) composition.EVD029
  • +20% accuracy lift from structured grounding (Atlan) composition.EVD020
  • -39% fewer agent tool calls with grounding (Atlan) composition.EVD020
  • 2× text-to-SQL accuracy with semantic model (Snowflake) composition.EVD019
  • ~50% agentic-AI projects to be cancelled by 2027 (Gartner) composition.EVD017
Ground an agent in the playground → Read the AI-grounding methodology →