ArchScript MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ArchScript MCP Review my architecture and list the rules it violates"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.

ArchScript is an architecture-as-code tool. You describe a system — domains, services, databases, queues, and deployments as cluster → server → Docker container — in a small language. ArchScript draws the diagram, checks the model against architecture rules and exports documentation, a PDF report, Mermaid, PlantUML and Terraform.
ArchScript for Mac includes an MCP server. Claude Desktop, Claude Code and other MCP clients can read the architectures you have open, change them with precise, structured edits, validate them and look at the rendered diagram. Every change appears live in the window and can be undone.
This repository contains the documentation, configuration examples, example models and a small open-source stdio bridge (
server/) for the ArchScript MCP server. ArchScript itself is a closed-source app, available on the Mac App Store. The browser Playground is free.
What Claude can do
Tool | What it does |
| Learns the language, element types and rules |
| Lists the open documents (id, name, validity) |
| Reads a document with line numbers and a structure outline |
| Creates a new architecture from ArchScript source |
| Replaces a document's source |
| Structured edits: add, rename, move and delete elements, relations and containers, and set properties, tags and colours |
| Returns diagnostics |
| Returns the review score, grade and rule findings |
| Returns the diagram as an image, so the assistant can see the result |
| Exports Mermaid, PlantUML, Markdown, JSON or Terraform |
The assistant doesn't rewrite your file. It sends small operations (for example
add_element into cluster:Production/server:app-01/container:api). The
ArchScript core turns those into minimal text edits, so comments and formatting
stay intact. After each change the assistant gets the validation result and
review score back, the same feedback loop a human architect uses.
Example prompts
"Read my architecture and tell me which rules it violates."
"Add a background worker that sends order emails. Put it in its own container on app-01 and connect the Order API to it."
"Create an event-driven architecture for a ticket shop with a payment provider and a read model."
"Export the current model as Terraform for AWS and list what wasn't mapped."
Related MCP server: jgs-magic-sysmlv1-mcp
Setup
Requirements: ArchScript for Mac (macOS 14+) with ArchScript Pro (14-day free trial).
Open ArchScript → Settings → AI and turn on Allow AI assistants (MCP).
Copy the snippet for your client from that screen. It already contains your personal access token.
Claude Desktop: one-click install (recommended)
Download archscript.mcpb from the latest release.
Double-click it, or drag it into Claude Desktop → Settings → Extensions.
Paste the access token from ArchScript → Settings → AI.
The extension contains no ArchScript code. It holds a manifest and a small
launcher script (mcpb/server/archscript-mcp)
that finds ArchScript for Mac and starts its bundled helper.
Claude Code
claude mcp add --transport http archscript http://127.0.0.1:47621/mcp \
--header "Authorization: Bearer <token from Settings → AI>"Claude Desktop (manual)
Add the snippet to claude_desktop_config.json, then restart Claude:
{
"mcpServers": {
"archscript": {
"command": "/Applications/ArchScript.app/Contents/MacOS/archscript-mcp",
"args": ["--token", "<token from Settings → AI>", "--port", "47621"]
}
}
}The bundled archscript-mcp helper bridges stdio to the app and starts
ArchScript when Claude needs it.
Other MCP clients
Any client that supports Streamable HTTP can connect to
http://127.0.0.1:47621/mcp with the header Authorization: Bearer <token>.
Clients that only support stdio can use the helper as shown above, or the
open-source Node bridge in server/ (Node 18+, no dependencies):
{
"mcpServers": {
"archscript": {
"command": "node",
"args": ["/path/to/archscript-mcp/server/index.js"],
"env": { "ARCHSCRIPT_TOKEN": "<token from Settings → AI>" }
}
}
}Privacy and security
Local only. The server listens on
127.0.0.1and is never reachable from the network.Token required. Every request needs your access token. You can regenerate it at any time in Settings → AI.
Off by default. Nothing runs until you turn it on.
No telemetry. ArchScript collects no data and sends nothing anywhere. Your AI client only sees what it reads through the tools.
The helper is sandboxed and only forwards stdio to the local server.
Screenshots
Visual editing | Review and rules |
|
|
Inspector | Element library |
|
|
A Docker deployment as code: cluster → servers → containers → services.

Example models
The examples/ folder contains ready-to-use architectures. Open one
in ArchScript for Mac, or paste it into the
Playground:
File | Architecture |
The smallest useful model | |
One domain with web app, API and database | |
Classic web application with boundaries | |
Services with their own databases and an event bus | |
Bounded contexts (domain-driven design) | |
Event-driven, CQRS-style flow | |
Production SaaS with gateway and identity provider | |
Docker Swarm cluster, servers and containers |
A taste of the language:
project DockerShop
cluster Production {
platform "Docker Swarm"
server "app-01" {
container api {
image "shop/api:1.4"
element service "Order API" as orders { tech "Rust"; owner "Orders Team" }
}
}
}
element frontend "Web Shop" as shop
relation shop -> orders : HTTPSThe full language reference is at archscript.dev/docs.
Feedback
Found a bug in the MCP integration or have an idea for a tool? Please open an issue. Feature requests for ArchScript itself are welcome too.
License
The documentation, configuration snippets, example models and the bridge in
server/ are available under the MIT License. ArchScript for Mac and the
ArchScript core are proprietary software.
Available Tools
10 toolsarchscript_guideArchScript language guideARead-only
The ArchScript language: syntax, element types, containers (domain, boundary, environment, cluster, server, container), relations, properties, rules, custom types and the edit operations. Read this before writing ArchScript.
Inputs: no named parameters.
create_documentCreate a documentB
Opens a new ArchScript window with the given source and returns its id and diagnostics.
Inputs: source (required).
edit_documentEdit a documentA
Applies structured edit operations in order (one undoable step). Each operation is an object with "op" and its fields: add_element {kind, name, container?} · add_relation {from, to, label?, bidirectional?} · set_property {element, key, value|null} · rename {element, name} · change_id {element, id} · change_kind {element, kind} · set_tags {element, tags[]} · move_to_container {element, container|null} · delete_element {element} · set_relation_label {relation, label|null} · set_relation_property {relation, key, value|null} · set_relation_direction {relation, direction: forward|reverse|both} · delete_relation {relation} · add_container {kind, name, container?} · rename_container {container, name} · delete_container {container} (keeps contents) · set_container_property {container, key, value|null} · set_container_color {container, color|null}. Elements are referenced by id, relations by index (see get_document), containers by key like "cluster:Prod/server:app-01".
Inputs: operations (required), document.
exportExportCRead-only
Generated output of a valid document: mermaid, plantuml, markdown (documentation), json (model) or terraform_aws / terraform_azure / terraform_gcp.
Inputs: format (required), document.
get_documentRead a documentARead-only
Source code (with line numbers), diagnostics, metrics and the structure (containers, elements, relations with their indices) of a document.
Inputs: document.
list_documentsList open documentsARead-only
Open ArchScript documents with id, file name, project name, validity and size. The first entry is the frontmost window.
Inputs: no named parameters.
render_diagramRender the diagramBRead-only
The document's diagram as a PNG image (the last valid state).
Inputs: document.
reviewArchitecture reviewBRead-only
Score, grade, metadata coverage, coupling and all rule findings of a document.
Inputs: document.
set_document_sourceReplace a document's sourceA
Replaces the whole source of a document (one undoable step in the user's window). Returns diagnostics.
Inputs: source (required), document.
validateValidate sourceARead-only
Compiles ArchScript source without opening or changing a document: diagnostics, metrics and review score.
Inputs: source (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Most tools have clearly distinct purposes, especially document CRUD versus export/render versus the guide. Some overlap exists: get_document, validate, and review all report diagnostics or metrics, and set_document_source versus edit_document both modify source, though their scopes differ.
Document operations mostly follow verb_noun naming such as list_documents, get_document, and create_document. However, export, validate, and review are bare verbs, while archscript_guide is an anomalous prefixed noun, making the overall convention mixed but still readable.
Ten tools is a well-scoped size for an architecture DSL server covering document lifecycle, editing, validation, review, export, rendering, and authoring guidance. No obvious tool feels redundant or trivially unnecessary.
The surface covers create/list/get/edit/validate/review/export/render plus a language guide, which is strong lifecycle coverage for ArchScript documents. Minor gaps include no document delete/close operation and no explicit open-by-file-path workflow.
Maintenance
Related MCP Connectors
Manage SysML2 projects; check, evaluate, verify, query, document and diagram models.
Render, verify, describe, and safely edit Mermaid diagrams through MCP.
Immutable agent diagrams: create, edit, validate and render versioned diagrams over MCP.
Read and edit the Revit 2024–2027 model on your PC. Needs the free Revit MCP add-in.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables querying and modifying ArchiMate enterprise architecture models from XML or Archi Tool files via REST API and MCP server, supporting multiple simultaneous data sources.19 npmMIT
- FlicenseCqualityBmaintenanceMCP bridge to live SysML v1 models in CATIA Magic / Cameo Systems Modeler (No Magic MSoSA). Lets Claude Code or any MCP agent query, audit, and edit MBSE models — blocks, requirements, diagrams, parametrics, traceability — over a local, air-gapped connection.1334-
- AlicenseNot gradedqualityBmaintenanceEnables planning and maintaining system architecture as typed, linted, live diagrams stored as files in the repo, with MCP tools for cataloging, creating, editing, linting, exporting, and viewing diagrams locally.8MIT
- FlicenseNot gradedqualityBmaintenanceProvides MCP tools to read, validate, propose, and review architecture models and UML diagrams in local workspaces.-



