Background On April 28, 2026, Anthropic launched Claude for Creative Work with an Autodesk Fusion connector built on the Model Context Protocol (MCP). Autodesk framed it as 'an early step in a broader shift toward a more open ecosystem where Autodesk software connects into the tools our customers already use.' The brief below identifies twelve workflow gaps in AutoCAD, Civil 3D, Subassembly Composer, Pipe Networks, and Inventor where the same MCP architecture would deliver immediate value. Each item is one sentence. A full report with documented pain points, sources, and feasibility analysis is available on request. Conclusions and Recommendations XEC requests consideration for any Anthropic Claude pilot programs Autodesk runs for DWG access, MCP connectors, or Civil 3D AI integration. XEC uses Claude in production today for catalog authoring (gravity XML, pressure SQLite, Inventor IPT scaffolding), Civil 3D template and style work, agency report formatting, and DWG analysis from extracted spreadsheets — and has reached the limits of what spreadsheet extraction or step by step workflow instructions can do. Section 1 — Tech Support and Drawing Health How many tech-support tickets does Autodesk field every day for basic DWG problems — files that won't open, crash on save, print wrong, or run slow? No tool today simply reads a DWG and reports its overall health; workarounds (spreadsheet exports, App Store utilities) are bounded by what AutoCAD's export and API surfaces expose. Direct DWG access through Open Design Alliance libraries — already available for out-of-process reading — changes that. DWG health analysis and cleanup — scan thousands of drawings for proxy entities, regapp residue, dictionary corruption, and other below-surface failure modes that don't appear in any spreadsheet, then dispatch Batch Save cleanup in parallel. External references (xrefs) — map every reference (attached, overlay, nested, image, PDF, data shortcut), explain in plain language what each does, flag broken or risky ones, and batch-repath including the types Reference Manager skips (PMP, OLE, hyperlinks, URL). Section 2 — Civil 3D Core: Templates, Data, and QA-QC Civil 3D's biggest pain points aren't in the design tools — they're in the infrastructure that holds projects together: templates that won't propagate, data shortcuts that break on a folder move, plan and profile sheets that drift out of sync, reports that never match the agency's format. Autodesk has shipped capable tools for each, but they require fluency across multiple subsystems and they don't talk to each other. Templates, styles, and label maintenance — diff project styles against a master DWT, identify drift, and import corrections under review; firm-wide style updates that take 1–2 weeks per upgrade cycle drop to hours. Or tell Claude a label you need and have it build it! Data shortcuts and DREF repair — use the Civil 3D 2020.6 RepairBrokenDRef API (largely unknown to CAD managers) to scan all consumer drawings and repair broken DREFs in batch, including profile DREFs that the native 'Repair Broken References' command doesn't handle. (Add Claude to the stand alone Data Shortcut Editor, and expand it’s capabilities) Plan and profile QA-QC cross-referencing — iterate every structure in plan, retrieve its station and rim/invert from the Civil 3D data model, locate the corresponding profile-view label, and flag discrepancies; no native, third-party, or App Store tool does this today. Agency report formatting and stormwater software output — generate agency-specific XSL templates from a sample PDF and cross-translate inputs between SWMM, HEC-RAS, and StormCAD. (Get the new Infor Drainage guys in on Claude) Batch Save Utility automation — wrap the existing headless LISP and .NET backend so every other workflow in this brief runs across thousands of drawings instead of one at a time. Grading Optimization — ingest agency drainage and slope specs (state, county, city, water district) and translate natural-language site descriptions into Grading Optimization criteria, turning an unused AEC Collection feature into a first-pass design tool. Assembly Section Automation — Be able to input a agency street or highway design details and have the ai create an assembly with all the subassemblies. Section 3 — Subassembly Composer Subassembly Composer is the gating artifact for creating special agency typical sections on any roadway project, and mastering it is a multi-month investment. PKT files are XML — open, parseable, and exactly the kind of structured format an LLM handles well. PKT authoring from agency typical sections — emit new PKTs from typical-section drawings or natural-language descriptions, enforce a firm-wide code namespace, and migrate older PKTs forward across versions; per-agency catalogs drop from multi-month senior-tech projects to days. Section 4 — Pipe Networks, Pressure Networks, and Inventor Civil 3D's Pipe and Pressure Network catalogs are arguably the platform's biggest unforced error — most firms avoid them because authoring a custom catalog is buried in three tools, two file formats, and a thirteen-year-old IdeaStation request that's still open. The April 2026 Fusion-Claude connector demonstrates LLM-driven parametric modeling, but Fusion-to-Inventor IPT export drops parametric history. An Inventor connector fixes that (I WANT THIS NOW, so I can create parts and not have to manually create them!!!) — and agency-specific catalogs are the gating artifact for DOT migrations off MicroStation. (One of the items I had Claude help me with is I put in a City manhole standard plan, and asked it to walk me through the steps of modeling it in Inventor) Gravity Pipe Network catalog authoring — author catalog XML and DWG part files directly from agency standard plate PDFs, resolving the 2013 IdeaStation request for IPT-to-Part-Builder import; custom municipal catalogs drop from weeks to days. Pressure Network catalog authoring — author SQLite catalogs from AWWA, ASTM, or manufacturer specification charts directly, eliminating the multi-thousand-dollar third-party catalog cost per material. Inventor parametric structure authoring — an Inventor connector analogous to the Fusion connector retains full parametric feature history and makes custom Civil 3D structures (manholes, inlets, vaults) accessible to engineers who design with them, not just specialists who model them. Section 5 — Strategic Context Two pieces of context make the case: the state DOT competitive landscape, and the MCP architecture Autodesk has already published. The first explains why this is urgent now. The second explains why it's low-risk. State DOT competitive landscape — roughly 47 of 50 U.S. state DOTs use Bentley OpenRoads or MicroStation; Bentley currently offers no Claude or MCP integration, so a Civil 3D connector that compresses agency catalog and subassembly authoring from months to days is a differentiator Bentley can't match in the short term. Existing MCP architecture — Autodesk Assistant is already in Fusion, Inventor, Moldflow, Vault, and Civil 3D 2027 (Tech Preview); reference MCP servers are public on Autodesk's own GitHub (aps-mcp-server-nodejs, aps-aecdm-mcp-dotnet, aps-sample-revit-mcp-tools-bundle), so this pitch is asking Autodesk to extend infrastructure that already ships, not build new.
Show More