Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Description from the maker (original English)
AutoCAD MCP/CLI for agents: native CAD, views, proxy reads, verified writes and saves.
These three are never merged into one score: opinions, measurements and popularity are different things.
We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.
Pick the AI app you use, then copy the settings.
Which app do you use?
Pick your app above. We remember it, so next time the button says “Install in …”.
Download the maker’s .mcpb file
The link goes to the maker’s own file; we do not host or repackage it. Open the file and Claude Desktop installs it.
Run this in your terminal
Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.
Your choice stays on this device in a cookie; no account needed.
These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 7, 2026
The registry name was verified with the GitHub account capskov, which owns the repo
We have not read this repo from GitHub yet
We have not read this repo from GitHub yet
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
The registry does not list a required key, though some servers ask you to sign in on first use.
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed –, and the latest version (1.2.0-beta.3) reached the registry 14 days ago.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need MCPB on your computer.
In the same category: Spec Driven Development, Remote Desktop Commander, Draugr
Servers in the same category from other makers, sorted by stars.
Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Security scanning for AI agents: SAST, SCA, secrets, IaC, DAST, ranked by real risk.
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
Current field release: 1.2.0-beta.3.18 adds local CTB controls, transient window PDF output and field reliability fixes. Validation and limits. Earlier milestones below remain historical.
Public runtime preview v1.2.0-beta.3. Proxy/compound geometry and text reads, bounded native views/captures, raster-path repair, rejection classification and polyline-width fixes. Reuses the accepted beta.3 binaries; scope, known capture limitations and upgrade notes.
Optional road interpretation: Agentic Map Tracing v0.1.0-beta.1. The Agent reasons about boundaries; Foundation supplies CAD execution. Install separately.
Foundation development supports the real task. Provide small, reliable, controlled CAD primitives and truthful feedback so the Agent can act and move forward. The Agent owns visual judgment, structure interpretation, example learning, pattern recognition and final review. Do not turn those judgments into programs, fitting algorithms or compulsory verification workflows. Do not add incidental features unless explicitly requested. When current primitives suffice and the Agent judges the result acceptable, continue the main task. Preserve drawing identity, source protection and no replay of unknown writes.
Local unit closeout: 1.2.0-local.4 carries prior bounded primitives, makes INSUNITS consistency optional for raw-coordinate plans and retains the legacy metres interface. This prerelease is published as v1.2.0-local.4; earlier releases remain available. See units, precision and paper typography; this is not blanket A–P certification.
Agent-first AutoCAD MCP — built for the execution loop, not just command calls. CAD Agent Foundation helps AI agents bind the right drawing, track each request, receive execution feedback, verify what was written or saved, and continue multi-step CAD work with confidence. Independent or third-party CAD Skills can reuse this execution foundation for their own professional workflows.
中文简介:面向 AI Agent 的 CAD 执行底座——不止调用命令,还关注图纸绑定、执行反馈、结果核验与安全续作,可与第三方专业 Skill 组合使用。
While CAD integrations connect AI to drawing commands and data, Foundation focuses on the agent workflow around each operation: knowing what was requested, what actually happened, and whether the next step is safe.
Public runtime preview v1.2.0-beta.3. Release and upgrade notes. Source, documentation and a manifest-bearing Windows runtime ZIP are provided in this preview. Build prerequisites and limits. The accepted local runtime v1.0.0-local is a separate baseline. Companion Skill previews are published independently: callout v0.3.0-rc.1 and distress v0.1.0-rc.1.
简体中文 · Quick start · Capabilities · Compatibility
Registered MCP server: io.github.capskov/cad-agent-foundation. MCPB installation and discovery · Registry metadata. Local Windows/AutoCAD prerequisites still apply; registration does not grant drawing access.
AutoCAD automation bridge for AI agents via MCP and CLI, using a native .NET API.
AI agents can inspect and operate a live AutoCAD drawing through structured native tools for bounded 2D geometry, text, hatches and callouts. The bridge uses MCP (Model Context Protocol) and CLI; drawing identity checks and write/save verification govern the supported operations.
CAD Agent Foundation gives CAD agents a bounded execution contract: explicitly bind a drawing, persist request ownership and execution state, inspect the original request after an uncertain reply, and verify tracked object facts after writes and saves. The current implementation integrates these mechanisms into a modified, pinned Bimwright DWG MCP host.
This Windows/AutoCAD bridge is maintained by capskov. The current preview is v1.2.0-beta.3, with an independently built runtime ZIP and scoped native evidence. The accepted v1.0.0-local baseline is retained; compatibility with every AutoCAD update is not claimed.
The project grew out of road and civil-engineering CAD work: creating boundaries, fills, quantities and linked callouts, then revising and saving them over a long agent session. A useful drawing command is only part of that workflow. The agent also needs to know which drawing it owns, whether a modification has already happened, what remains unverified, and whether another attempt would duplicate it.
The design focuses on those decisions. Durable records, drawing identity and native readback add overhead; no comparative speed, token-cost or productivity advantage has been established. Engineering judgment and domain-specific layout rules remain in separate Skills.
Download v1.2.0-beta.3 runtime, verify SHA256SUMS.txt and extract the complete folder. Requires licensed compatible Windows x64 AutoCAD 2026, .NET 8 Desktop Runtime and Python 3.11+. No Autodesk assemblies or SHX fonts are redistributed. Installation · 25-operation catalog · New validation.
New in this preview: installed-font text styles, dimension styles, native linear/aligned/radius/diameter DIMENSION, layer state management, guarded closed-boundary/hatch continuation, linked callout geometry/format, and same-drawing rotate/scale/copy of supported complete groups. It preserves original-request reconciliation and drawing/object protection.
Maturity: the bounded local runtime has native acceptance evidence; The release provides a runtime ZIP with dependency licenses, file hashes and plugin MVID; prerequisites remain separately installed. Support follows the actually exposed profile.
| Category | Current capabilities | Status / limits |
|---|---|---|
| Basic drawing | New layer, line, circle, 2D polyline and associative hatch on an earlier new boundary | Specified local workflows verified; not arbitrary editing/all hatch cases. |
| Text and callouts | Single-line DBText, multiline MText, ordinary LEADER/MText/yellow filled-dot group | Creation verified; complete domain styles/placement belong to Skill. |
| Edits / object management | Digest-guarded linked-text update and complete callout-group move/archive | Specified group workflow verified; not every entity branch individually tested. |
| Execution / delivery | Drawing binding, original-request queries, duplication guards and whole-document save/disk readback | Local workflows verified; no automatic unknown-write replay/unlock. |
| Reads / helpers | Entity properties/digests, layer list, filtered paging and static Lisp inspection | Entity readback verified; some other implemented routes lack current-version individual end-to-end proof. |
| Additional bounded operations | Four native DIMENSION types, named matching styles, layer states, closed-boundary/callout updates, rotate/scale/copy | See the 25-operation catalog for exact guards and scoped native evidence. |
| Not exposed | MLEADER, shared-style overwrite, arbitrary object editing, complex Boolean/holes, layout/plot/xref edits | No blanket coverage of every native object or upstream command. |
Full interfaces, invocation and per-item verification status. Foundation supplies reusable native operations, execution constraints, status and object management. Independently maintained Skills supply domain rules, style parameters, placement algorithms and workflows through actual supported Foundation interfaces. Missing native/backend compatibility is not supplied merely by installing a Skill.
Bind the intended drawing and verify process/session/plugin identity. Persist a stable request before submission; after an uncertain reply, query that original request. Existing edits need fresh native digests and complete associations. Confirm effects, independent readback and fence release separately; save covers the whole working drawing. See architecture and versioned interface catalog.
| Task | Required components | Versions and limits |
|---|---|---|
| Supported basic CAD operations | cad-agent-foundation only | Local baseline v1.0.0-local; public source preview v1.1.0-rc.1, with a reviewed runtime ZIP; prerequisites are separate. Bounded 2D operations; local.4 raw writes use original drawing values; unit metadata checks are optional; recorded Windows/AutoCAD 2026 host only. Neither Skill is required. |
| This project's callout style and placement | Foundation + cad-callout-pair-layout | Public callout Skill v0.3.0-rc.1 (existing rules v0.3). Metre-based road defaults; Native style/pose interfaces and a bounded synthetic unchanged-Skill integration are verified in v1.1; full real-project/background coverage remains per task. Check backend coverage before writing. |
| This project's distress/demolition workflow | Foundation + cad-distress-annotation + cad-callout-pair-layout | Public distress rules preview v0.1.0-rc.1. Requires engineering evidence, classifications and the callout rules; full workflow needs capabilities outside the current Foundation profile and is not certified on these three components alone. |
The two Skills are optional for basic operations. Their rules are not built into Foundation; installing them does not add missing native operations. Foundation source and both independent Skill previews are public. Installation entries, versions and Agent checks.
The commands below use the extracted runtime ZIP. Source checkout alone requires build/packaging first.
foundation.cmd install --workspace "C:\CAD-work\demo01"
foundation.cmd start --workspace "C:\CAD-work\demo01" --source "C:\drawings\input.dwg" --write foundation.cmd doctor --workspace "C:\CAD-work\demo01" foundation.cmd apply --workspace "C:\CAD-work\demo01" --plan "plan.json" foundation.cmd save --workspace "C:\CAD-work\demo01" --handles RETURNED_HANDLES foundation.cmd stop --workspace "C:\CAD-work\demo01"
Omit --write for a read-only session. start creates a working copy; it does not overwrite the source drawing. Save affects the entire working drawing, including other unsaved edits in that instance. Read the quick start for request handling and a small synthetic plan.
The local release was checked on two different drawing sources, with different coordinates and handles. Each workflow created ten new objects, read them independently, updated linked text, moved the complete callout group, archived it, saved and closed the drawing. Fresh CAD sessions then read the saved objects with matching native digests. The CLI/MCP entry and local package extraction/installation were also checked.
These are bounded workflow results on a recorded Windows AutoCAD 2026 environment. They do not prove universal drawing correctness, production-wide reliability, compatibility with every patch, or a general CAD engineering agent. The validation summary separates native evidence from offline tests and records unresolved historical outcomes honestly.
Plans write bounded 2D model-space geometry. Coordinates, lengths and annotation sizes are expressed in drawing units. WCS is the CAD coordinate reference; i
Only the beginning is shown; read the rest in the repo