Architecture firms

AutoCAD plans & 3D in Claude

Massive 2D/3D plans, precise context, optimized in Claude.

integer-precision coordinates

records + graph

one plan, two shapes

NBC / CSA

conventions, cited

in-chat

answers with sources

How it fits together


   ┌───────────┐  export   ┌──────────────┐   recall + optimize   ┌─────────┐
   │  AutoCAD  │ ────────► │  Memoception │ ◄───────────────────► │ Claude  │
   │  DWG/IFC  │  DXF/IFC  │   records +  │      (MCP connector)  └─────────┘
   └───────────┘           │   adjacency  │
                           │   graph      │  ── conventions (NBC/CSA) tagged
                           └──────┬───────┘     with code_ref
                        Memoception-1.5 distills → decisions, dims, constraints

The challenge

AutoCAD projects carry enormous 2D plans and 3D models, plus building codes and specs. Pulling precise context out of them — a corridor width, an adjacency, a convention — is slow and error-prone, and nothing survives into the design conversation.

How Memoception solves it

Store large plans and 3D

DWG/DXF and IFC exports are custom-stored with integer-millimetre precision — rooms as records, adjacencies as a graph — so a plan becomes both a schedule and a spatial graph.

Get context from any format

Specs, code excerpts, and drawings ingest across formats; a large memory of Canadian (NBC/CSA) conventions sits alongside, tagged with its code references.

Optimize with Memoception-1.5

Summarize distills a sprawling project into its strongest form — the load-bearing decisions, dimensions, and constraints — so the model reasons over signal, not noise.

Use it inside Claude

Through the MCP connector the architect asks Claude about a plan — 'what's the accessible corridor width here and where did it come from?' — and gets the stored measurement with its code_ref.

Built with

custom/store (json + neo4j)custom/recall (table, graph)summarize (Memoception-1.5)MCP connector

Build this →Request a custom integration