Skip to main content

Analyze your data

Ask questions about your live infrastructure in plain language and get an answer back, without writing a query. Use this to check what depends on a circuit before a maintenance window, find the devices missing a platform, or see where the deployed state has drifted from the design.

You do not have to be the engineer who built the model. Your assistant reads your schema through the Infrahub MCP server, so it works out which kinds and relationships to follow from the question you asked.

Prerequisite

Requires the Infrahub MCP server connected to a running instance. See Set up your AI assistant.

To ask questions and change nothing, a read-only deployment is enough: with INFRAHUB_MCP_READ_ONLY=true the write tools are not offered to your assistant, and GraphQL mutations are rejected.

Querying​

Ask single questions — find, list, count:

List devices without a primary IP address.
Which prefixes in this pool are unallocated?

Ask relationship questions the same way:

How are these two devices connected?

Behind those answers, your assistant uses the MCP server's tools directly: get_nodes and search_nodes for straightforward reads, query_graphql for anything relationship-heavy, and find_paths and find_reachable for graph traversal.

For the same relationship questions answered by hand, see Analyze dependencies and Query traversal with GraphQL.

Analysis​

These six categories cross several domains at once, which is what makes them awkward to express as a single query:

  • Compliance — "Which devices break our naming standard?"
  • Service impact — "Which services depend on this circuit?"
  • Maintenance windows — "What is affected if this device goes down Saturday?"
  • Drift detection — "Where does the realized state differ from the design?"
  • Capacity planning — "Which resource pools are near their limits?"
  • Inventory gaps — "Which devices are missing a platform or location?"

When this is the wrong tool​

Each of these answers the question for your data as it is now. Where a rule should hold for every future change, it belongs in a check that runs on every Proposed Change instead — see Checks & Validation. Compliance questions are the common case: asking which devices break a naming standard tells you about today, and a check keeps it true.