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., "@ART-CLI Create a geometric logo in VectorCraft, save the editable project, and export SVG and PNG."
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.
One on-demand bridge between AI agents and 13 ArtCraft application families. Create and edit real documents, discover native capabilities, save editable projects and export deliverables — through a native Hermes plugin, a CLI, or a stdio MCP server.
Built by kosecom. Independent integration; not an official ArtCraft or Hermes product. Applications and their licenses are separate. Third-party notice →
Download · Give it to your agent · Setup · Deutsch · Technical reference
Download the release ZIP, extract it, and tell your agent:
Read
START-HERE.md. Inspect this package, check my native ArtCraft CLIs, and install ART-CLI into the profile I choose. Use the Hermes plugin if available; otherwise configure stdio MCP or the CLI. Do not download software or change app permissions without asking. Verify the connection and show me which apps are ready.
You do not need all 13 apps. Use the installed subset. Missing apps are reported, not silently replaced with another app or host.
Requires native Windows or macOS and Python 3.10+. For Hermes, use its Python environment and a compatible native-plugin runtime. On macOS the interpreter is often called python3.
# Read-only check: no apps launched, no downloads, no installation.
python setup_art.py --home "<absolute-target-home>" --agent hermes --check
# Install into the same explicitly selected home.
python setup_art.py --home "<absolute-target-home>" --agent hermes --installChoose --agent mcp for a generated stdio configuration or --agent cli for direct commands. Follow the setup report for Hermes activation and start a new agent session. Never select another profile by accident.
GUI app installed ≠ native CLI installed. macOS may need separate official CLI assets. Setup reports missing components; it never silently downloads or installs vendor software. Dependency setup and activation →
Related MCP server: Adobe Creative MCP
Work | Applications | Examples |
Graphics & photography | VectorCraft, PhotoCraft, LightCraft, FigCraft | Vector logos, layered edits, photo adjustments, layouts |
Page design & documents | DesignCraft, WordCraft, PdfCraft | Page layouts, text documents, supported PDF edits |
Presentations & data | DeckCraft, GridCraft | Slide decks, spreadsheets, formulas, tabular exports |
Motion, video & audio | EffectCraft, FilmCraft, SoundCraft | Compositions, timelines, offline processing, supported exports |
Technical drawing | CADCraft | Drawings, geometry, supported CAD exports |
Capabilities and formats come from your installed app version, not invented commands. Search and describe the exact tool before execution.
Try: “Create a geometric logo in VectorCraft. Save an editable project and export SVG and PNG. Reopen the project and check the export.”
Or: “Inspect this workbook, update these cells, save to a new file and verify the values.” These are task examples, not a guarantee for every version, codec or export option.
Hermes: eight progressive tools, not hundreds of permanently loaded schemas.
Tool | Purpose |
| App inventory; explicit |
| Paginated tool and editor-command discovery |
| Exact schemas, arguments and invocation mapping |
| One native action |
| Ordered actions in the same job; prior-result references |
| Usage and app guidance |
| Original native CLI arguments and optional stdin |
| Native MCP resources/prompts or configured GUI control |
CLI: the same backend without a Hermes chat. From an extracted package:
python art.py --home "<absolute-target-home>" catalog
python art.py --home "<absolute-target-home>" doctor
python art.py --home "<absolute-target-home>" refresh --app vectorcraft
python art.py --home "<absolute-target-home>" search --app vectorcraft --scope tools
python art.py --home "<absolute-target-home>" guide --app vectorcraftMCP: python art.py --home "<absolute-target-home>" mcp exposes one generic art tool. Setup generates absolute interpreter/script paths. Clients must support stdio MCP and answer server ping requests. No public HTTP listener is created. The protocol is tested; compatibility with every agent UI is not claimed.
Lazy by default: registration and cached discovery do not launch apps.
Real files: explicitly save/export, then reopen and verify the artifact.
Owned-process cleanup: Windows Job Objects and macOS process groups isolate job-owned children. Pre-existing GUIs are not swept up by app-name kills.
No silent edit retries: interrupted mutations may be partial. Batches are not transactions.
No approval invention: native callbacks are not auto-granted. File roots, permissions and GUI credentials remain operator decisions.
No private machine state in the ZIP: no vendor binaries, credentials, personal paths, live profiles or SSH routes.
Headless jobs use the actual native engine but do not automatically edit your visible document or populate “recent projects.” Open the saved native project to continue manually. GUI attachment and remote routing are separately configured advanced features. GUI control → · Architecture →
Native Windows and macOS tests cover plugin discovery, app reads, both remote directions, process lifecycle, malformed frames, cancellation, package integrity and installer behavior. An audited snapshot contained 728 native tool schemas and 6,704 editor entries across 13 apps per host.
Schema coverage is not execution coverage. These figures do not mean every editor action, GUI gesture or export combination was executed. This release includes fixes for malformed cancellation and heartbeat-response IDs. Verification and limitations →
Upgrade: add --upgrade to the setup installation command. Managed previous installs are backed up.
Removal: disable the Hermes plugin in its owning profile first, then python install.py uninstall --home "<absolute-target-home>". Apps are not removed; retained data is explained in the setup guide.
Report a bug with OS, Python/agent version, app CLI version, command and redacted errors. Never attach tokens, private documents or personal configurations.
The banner and explanation graphics were created through ART-CLI with the native VectorCraft engine. Editable sources are distributed separately with the release. They demonstrate this workflow, not a GUI session.
MIT-licensed integration code. Third-party apps and assets retain their respective licenses.
One ArtCraft entry. Start with catalog/guide, then search and describe for exact native schemas. execute/batch start only selected native apps; explicit save/export required. All raw vendor CLI functions through native; resources/prompts through rpc. Not automatic live GUI control.
| Name | Required | Description | Default |
|---|---|---|---|
| arguments | No | ||
| operation | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full behavioral burden. It does disclose real constraints: execute/batch only start selected native apps, explicit save/export is required, and there is no automatic live GUI control. What is missing is side-effect, permission, and state-persistence detail for a tool that can launch applications.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The text is a run of telegraphic, semicolon-separated fragments with no subject in most clauses, front-loaded with the least informative sentence ('One ArtCraft entry'). Density is high but readability is low; the ordering does not help an agent parse the routing rules.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
This is an 11-operation dispatcher with a nested arguments object, 0% parameter coverage, no annotations, and no output schema, yet the description never enumerates the operations or explains the argument structure. It omits doctor, refresh, and control entirely, leaving an agent unable to call most of the surface correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description must compensate and largely does not. It mentions a handful of the 11 enum values (catalog, guide, search, describe, execute, batch, native, rpc) but never defines any of them, and the generic nested 'arguments' object is not explained at all despite being where operation-specific payloads must go.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The tool name 'art' is opaque and the opening line 'One ArtCraft entry' largely restates it, but the following sentences hint at the real function: a single gateway exposing catalog/search/describe/execute/batch/native/rpc operations. There is no explicit verb+resource statement such as 'invoke ArtCraft native app operations', so the agent must infer the purpose from the operation hints rather than read it.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It does prescribe a workflow: start with catalog/guide, then search and describe to get native schemas, and use execute/batch only for selected native apps. It also routes raw vendor CLI through native and resources/prompts through rpc. However, operations like doctor, refresh, and control get no guidance, and there are no explicit when-not-use conditions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.2.0artScored across 1 tool
With only one tool there is literally no opportunity for misselection, so the set cannot be ambiguous across tools. However, the single tool's own purpose is dense and opaque (catalog/guide/search/describe/execute/batch/rpc/native all funneled through one entry point), which makes its scope hard to pin down.
There is no set of names to compare, so no inconsistency exists, but 'art' is a bare noun that establishes no verb_noun convention and is vague about what it does. It is readable but not descriptive of the underlying dispatch behavior.
A single tool is far too few for a server that apparently fronts an entire ArtCraft/vendor CLI with catalog, search, describe, execute, batch, and rpc surfaces. Everything is collapsed into one catch-all entry, forcing the agent to guess subcommands rather than choose a tool.
The description implies coverage of all native CLI functions plus a resources/prompts rpc bridge, so capability may exist behind the single tool. But the surface itself exposes no discrete, discoverable operations, leaving the agent dependent on self-describing catalog/guide steps to reach coverage.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Automate 1,000+ services from any MCP-compatible AI agent: build Applets, run actions and queries.
Agent-native notes, tasks, dev-docs, vaults, sync & handoffs. MCP + OpenAPI dual surface.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
