--- title: "How To Export DXF, DWG, and SketchUp Site Files Directly From An AI Chat" description: "Ask Claude or ChatGPT for site intelligence and get back a georeferenced DXF, DWG, SKP, or IFC file in the same conversation — no manual redraw, no portal hopping, no GIS software." canonical: https://atlasly.app/blog/export-dxf-dwg-site-analysis-from-ai-chat published: 2026-04-19 modified: 2026-04-19 primary_keyword: "export DXF DWG from AI chat" target_query: "how to export DXF from AI chat or Claude / ChatGPT" intent: commercial --- # How To Export DXF, DWG, and SketchUp Site Files Directly From An AI Chat > Ask Claude or ChatGPT for site intelligence and get back a georeferenced DXF, DWG, SKP, or IFC file in the same conversation — no manual redraw, no portal hopping, no GIS software. ## Quick Answer You can export DXF, DWG, SKP (SketchUp), or IFC site files directly from an AI chat by connecting Atlasly's MCP server (https://mcp.atlasly.app/mcp) to Claude or ChatGPT, then asking the AI to run a site package on an address. The chat returns a signed download link to the georeferenced CAD/BIM file, which opens in AutoCAD, Revit, SketchUp, or Rhino with no manual cleanup. ## Introduction Architects spend hours every week downloading CAD base files from planning portals, converting formats, fixing coordinate systems, and redrawing site context. With an AI chat connected to Atlasly, that entire workflow collapses into a single conversation. This article shows how to ask Claude or ChatGPT for a full site analysis and walk away with a georeferenced DXF, DWG, SKP, OBJ, or IFC file — all without leaving the chat window. It works because Atlasly's MCP exposes the full 17-step site-intelligence pipeline as tools the AI can invoke, and every export is produced by the same Atlasly runtime that powers the web app. ## Why exporting CAD files from AI chat is useful The conventional workflow for a new project looks something like this: 1. Open Ordnance Survey or planning portal. 2. Download a PDF or DXF of the area. 3. Open the file in AutoCAD or Revit. 4. Realise the coordinate system is wrong. 5. Redraw layers manually, because the source file has no semantic layer structure. 6. Spend half a day before design work begins. The AI-chat flow replaces all six steps with one sentence. The AI understands the address, asks Atlasly to run the pipeline, and hands back a file already aligned, layered, and ready to drop into your project. Beyond speed, the chat approach gives you context — the AI can also explain flood zones, heritage constraints, and planning history in the same conversation where it delivered the file. This is impossible in a portal-download workflow. ## The export formats Atlasly supports from MCP Atlasly exposes 14 file formats, accessible via the `run_site_package` tool through the MCP: **CAD** - DXF — AutoCAD, Illustrator, Rhino, QGIS - DWG — AutoCAD, Civil 3D - SKP — SketchUp **3D / BIM** - GLB — glTF binary (web, VR, AR) - OBJ — universal polygon format - FBX — animation-capable, Maya / 3ds Max - STL — 3D printing - Collada (DAE) — legacy SketchUp pipelines - IFC — BIM-ready for Revit, ArchiCAD, Tekla **GIS** - GeoJSON — web mapping - Shapefile ZIP — QGIS, ArcGIS - SVG — Illustrator, presentations **Data** - CSV tables - PDF site intelligence report All exports preserve EPSG:4326 georeferencing and use named layers (for CAD) so files open at the correct coordinates in your destination software. ## Step-by-step: exporting a DXF from Claude **Step 1.** Install the Atlasly MCP in Claude by pasting `https://mcp.atlasly.app/mcp` into Settings → Connectors. Full install guide at https://atlasly.app/blog/how-to-use-atlasly-in-claude-chatgpt. **Step 2.** Sign in to Atlasly (the CAD export tool is authenticated). Claude will OAuth you on first use. **Step 3.** In Claude, ask: *"Using Atlasly, run a full site package for 10 Downing Street, London SW1A 2AA, then give me the DXF download link."* **Step 4.** Claude calls `run_site_package` which starts the 17-step site-intelligence pipeline asynchronously. You get a job ID and progress updates. **Step 5.** After roughly 60 seconds, Claude returns a message containing a signed download URL for the DXF (plus options for DWG, SKP, IFC, GLB, and PDF). **Step 6.** Click the link. The file downloads to your local machine. Open it in AutoCAD — the site is correctly positioned, layers are named `BUILDINGS`, `FLOOD_ZONE_3A`, `LISTED_BUILDINGS`, `TREES`, etc. **Step 7.** Continue design work. Zero manual cleanup. ## What the georeferencing actually does Most CAD files downloaded from UK portals arrive in one of two broken states: - **Wrong coordinate system.** OS data is usually OSGB36 / EPSG:27700. Imported into a modern AutoCAD project expecting EPSG:4326, the site ends up somewhere in the Atlantic. - **No coordinate system at all.** Drawn around a local origin point that means nothing relative to the real world. Atlasly exports use EPSG:4326 (WGS84) as the canonical CRS and embed the transformation metadata in the DXF header. AutoCAD and Civil 3D pick this up automatically via the `MAPCONNECT` command or direct import with the GIS workflow set. Rhino picks it up through the OpenNURBS geo-referencing extensions. If your project uses a different CRS (e.g. OS National Grid for a UK site being delivered into a council project), Atlasly's file includes both the original WGS84 coordinates and the OS equivalents in comments so your reprojection is lossless. ## Named CAD layers — what you get The DXF / DWG export uses the following named-layer structure, matching common AEC conventions so it drops into a template without re-layering: - `SITE_BOUNDARY` - `BUILDINGS_EXISTING` - `BUILDINGS_LISTED_GRADE_I` - `BUILDINGS_LISTED_GRADE_II_STAR` - `BUILDINGS_LISTED_GRADE_II` - `ROADS_PRIMARY`, `ROADS_SECONDARY`, `PATHS` - `FLOOD_ZONE_2`, `FLOOD_ZONE_3A`, `FLOOD_ZONE_3B` - `CONSERVATION_AREA`, `SCHEDULED_MONUMENT`, `REGISTERED_PARK` - `ECOLOGY_SSSI`, `ECOLOGY_ANCIENT_WOODLAND` - `TREES_PROTECTED`, `TREES_EXISTING` - `UTILITIES_KNOWN` (where data available) - `CONTEXT_500M`, `CONTEXT_1KM`, `CONTEXT_2KM` For BIM (IFC), the same semantic structure is exposed as IfcSite containing IfcBuildingProxy elements tagged with Pset_Atlasly property sets. This means Revit schedules can filter by constraint type out of the box. ## Exporting from ChatGPT The flow is identical once you have the Atlasly connector installed in ChatGPT. Ask: *"Using Atlasly, run a site package for [address] and give me DXF, IFC, and PDF."* ChatGPT will call `run_site_package` through its custom-tools framework and return a message with the three download links. File generation happens server-side at Atlasly; ChatGPT just surfaces the signed URLs. One practical difference: ChatGPT's async tool-call UX sometimes truncates long responses more aggressively than Claude. If you only see one format link returned, ask "give me the other export links" and ChatGPT will pull them from the already-completed job. ## Costs, limits, and quotas **Public MCP tools (including `analyze_site`, `get_flood_risk`):** free, anonymous, 5 calls per 30 days per anonymous fingerprint. **Authenticated tools including `run_site_package`:** require a free Atlasly account. On the free Starter plan, you get 5 site analyses per month and PDF / image exports. CAD / BIM export formats (DXF, DWG, SKP, IFC, GLB) are on the Professional plan (£14.99/mo) which also lifts the analysis limit to unlimited. **API call quotas (if you invoke tools programmatically rather than via chat):** Professional 100 calls/month, Teams 1,000 calls/month. The MCP itself does not charge per call — quota is counted against your Atlasly plan the same as if you ran the analysis in the web app. ## From Practice A small practice we work with tested this end-to-end on an acquisition screening run. They had 12 potential sites to compare over a weekend. Previously this meant 12 separate portal-and-CAD dances, probably three full days of work. Using Claude + Atlasly MCP, they ran all 12 in one Sunday afternoon, got DXFs for every site, and had a comparison spreadsheet ready for the Monday partner meeting. The director sent us a note Monday morning with the subject line 'this is unreasonable'. Not the feedback we were expecting. ## Frequently Asked Questions **Can I export DXF from ChatGPT?** Yes. Install the Atlasly MCP connector (https://mcp.atlasly.app/mcp) in ChatGPT's Connectors settings, sign in to Atlasly, and ask the AI to run a site package and return a DXF link. Paid ChatGPT plans with the Connectors feature are required. **What AutoCAD / Revit version is required?** Any recent version. DXF output is compatible with AutoCAD 2010+. DWG with 2018+. IFC is IFC4, compatible with Revit 2022+, ArchiCAD 25+, and Tekla Structures 2022+. **Are Atlasly CAD exports georeferenced?** Yes. Files use EPSG:4326 (WGS84) as the canonical CRS with transformation metadata embedded. AutoCAD and Civil 3D pick this up automatically. OS National Grid (EPSG:27700) coordinates are included in file comments for UK projects needing OS reprojection. **Do I need to redraw anything after the export?** No. Named layers, correct units (metres), and georeferencing are all preserved. Files drop into your template without manual cleanup. **Do CAD exports cost extra on top of my AI subscription?** No additional fee from Atlasly. CAD / BIM formats are included on the Atlasly Professional plan (£14.99/month). Your AI subscription (Claude Pro, ChatGPT Plus, etc.) is separate and unchanged. **Can I get an IFC file for Revit from the AI chat?** Yes. Ask the AI to run a site package and specify IFC. The returned file is IFC4 schema, imports directly into Revit / ArchiCAD / Tekla. IfcSite geometry, building footprints, and constraint overlays are all included. **How accurate is the CAD geometry for formal submissions?** Building footprints are from Microsoft Global Buildings + OSM + authoritative sources. Accuracy is ±1-3 metres for most UK urban sites. Terrain is Mapbox Terrain-RGB, ±1-5 metres. This is accurate enough for feasibility, massing, and shadow studies. For construction or planning-submission-grade survey accuracy, still commission a topographical survey and LiDAR where required. ## Conclusion Exporting georeferenced DXF, DWG, SKP, or IFC site files directly from an AI chat is not a novelty — it is now the fastest path from address to design-ready base geometry in UK architectural practice. One link pasted into Claude or ChatGPT, one plain-English question, and you have the file. Install free — https://mcp.atlasly.app/mcp in Claude Settings → Connectors. Try it on a live site this week. ## Related Reading - https://atlasly.app/blog/how-to-use-atlasly-in-claude-chatgpt - https://atlasly.app/blog/export-site-analysis-to-autocad-and-revit-without-redrawing-the-site - https://atlasly.app/blog/atlasly-mcp-tools-reference-site-analysis - https://atlasly.app/blog/free-ai-site-analysis-tool-for-architects --- Source: https://atlasly.app/blog/export-dxf-dwg-site-analysis-from-ai-chat Platform: Atlasly — AI site intelligence for architects, engineers, and urban planners. https://atlasly.app