Codes

It knows the rules.

State the project — the jurisdiction, the occupancy, where the material goes, sprinklered or not — and read what each material does about the building code there. Every finding cites the requirement it was read against.

52 jurisdictions on the reference · 6130 seeded requirements, each carrying its citation

How a material gets read against a code.

01

The reference

Model codes as adopted, jurisdiction by jurisdiction — what the code actually asks of a material in that place, for that occupancy, in that location in the building.

02

The reading

Each material's typed values are held against the requirement that applies — the same canonical fire-classification tokens the rest of the graph runs on.

03

The finding

Meets, falls short, or nothing on file. Every material stays on the board — a gap in the record is a finding too, never a silent drop.

Honesty

Advice, and it says so on every finding.

Advisory — model code as adopted, not a legal determination. Always verify with the authority having jurisdiction.

That sentence ships with the data — it is the registry’s own constant, attached to every advisory the graph serves, on every surface. Coverage is bounded by the reference: a jurisdiction the reference doesn’t hold returns nothing rather than a guess.

For machines

An agent can ask about the rules too.

The advisory rides the MCP surface as typed tools: discover what the reference covers, then read a material against a stated project. The caveat travels in the response.

Explore the developer surface →
MCP
claude mcp add --transport http materialgraph \
  https://beta.materialgraph.com/mcp \
  --header "Authorization: Bearer mg_read_..."

# then, from any MCP-native agent:
#   list-code-contexts    — the jurisdictions and dimensions the advisory covers
#   resolve-code-advisory — a material read against a stated project