sketchy-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| SKETCHY_EDITOR | No | Editor for file links. Allowed values: vscode, cursor, windsurf, vscode-insiders, zed. | vscode |
| SKETCHY_PHOTOS | No | Set to off to disable photo search and labeling. | on |
| SKETCHY_DATA_DIR | No | Directory for models, cached photos and the local SQLite database. | ~/.sketchy |
| SKETCHY_EXPORT_DIR | No | Folder the save dialog opens in, and where exports go when no dialog is available. | ~/Downloads |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Sketchy draws hand-sketched, interactive diagrams in the chat. Call sketch once with the whole scene; Sketchy computes all positions. Then explain in 2-4 sentences instead of describing the drawing.
Patterns: flow (left-to-right chain + one step per stage), structure (groups for areas, dir down for hierarchies), comparison (one group per option, parallel nodes, !accent on the key difference), metaphor (real vs everyday group, dashed links). 3-12 nodes, labels 1-4 words, details after ~. Icons sparingly; !accent on 1-2 elements. 3-7 steps.
Follow-ups: to change an earlier drawing, send the whole scene again with the same ids; elements that stay keep their places. Circuits: search_parts for types and pins, then part and wire lines (red power, black GND, a resistor with every LED). Photos: find_image, locate the regions, then sketch with an image line and label lines for the regions. Keep the attribution.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| sketchA | Draw a whole scene and show it. One element per line: title | subtitle | dir right|down group [in ] [@icon] [(cylinder|person|note|circle|rounded|text)] [in ] [!accent|!muted] [file=] [~ hover text] -> [label] [!accent|!muted] (--> dashed, <-> both ends) label [.] (photo region with spaces: label .: ) image step : | <ids to highlight; edges as a->b> part [key=value] [caption] (types and pins: search_parts) wire <part.pin> -> <part.pin> [red|black|blue|green|yellow|orange|purple|gray|white] Icons: @globe, @database, @server (Lucide) or @simple-icons:react. Ids: letters, digits, _ or -. |
| search_partsA | Search the part library (boards, sensors, displays, LEDs, …). With an exact type it lists the part's pins. |
| find_imageA | Search freely licensed photos (Wikimedia Commons, Openverse; CC0, public domain, CC BY, CC BY-SA). Returns candidates with ids and thumbnails. Prefer clear, frontal photos of a single object. |
| locateA | Find named regions in a photo with the local vision model, e.g. ['USB port', 'reset button', 'A0']. Printed text is matched exactly; describe other parts visually ('silver USB connector'). Returns a preview with the boxes drawn so you can check them: rephrase queries that missed, skip wrong boxes, and if a box marks a different part than asked, you may still use its region name with corrected label text. |
| choose_modelA | Show the user a picker to choose and download the local vision model used by locate (one-time setup, can be changed later). |
| download_modelC | Start downloading a photo-labeling model. Used by the model picker. |
| model_statusB | Report the state of the photo-labeling model. Used by the model picker. |
| get_viewC | Returns the data behind a Sketchy view. Used by the viewer. |
| open_linkC | Opens a file from the drawing in the user's editor. Used by the viewer. |
| save_exportC | Saves an SVG, PNG or Excalidraw export of the drawing. Used by the viewer. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| explain-structure | Labeled drawing of what a thing is made of. |
| explain-flow | Flow diagram with a guided step-by-step walkthrough. |
| compare | Side-by-side comparison that highlights the key differences. |
| metaphor | Maps an unfamiliar concept onto something familiar. |
| explain-repo | Architecture diagram of a codebase; clicking a box opens the file. |
| explain-circuit | Wiring diagram generated from Arduino code. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| Sketchy viewer | Renders Sketchy scenes. |
TDQS
Scored across 10 tools
Core tools have distinct purposes: sketch draws, search_parts looks up hardware, find_image/locate handle photos, and save_export writes files. The only friction is that six of ten tools are labeled 'used by the viewer' or 'used by the model picker' plumbing, which could leave an agent unsure whether they are meant to be called directly. The model trio (choose/download/status) is clearly differentiated from each other.
Most names follow a clean verb_noun pattern (search_parts, find_image, choose_model, download_model, get_view, open_link, save_export). Deviations are minor: 'sketch' and 'locate' are bare verbs and 'model_status' is a noun phrase, but all remain readable and unambiguous.
Ten tools is a sensible, well-scoped set for a diagramming server. Slightly top-heavy, since six of the ten are support plumbing (viewer and model management) rather than drawing capability, but nothing feels redundant.
The surface covers creation (sketch), asset sourcing (search_parts, find_image, locate) and export (save_export), but lacks any explicit edit, delete, or element-removal operation, so incremental changes require re-emitting the whole scene. Current-state inspection is nominally available via get_view, yet it is described as viewer-internal, leaving agents without a clean read-back path.