Studiolo 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., "@Studiolo MCP Find the screenshots I saved last week and show me what's in them."
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.
Studiolo MCP
Let your AI assistant search, look at, and organize everything you save in Studiolo.
Studiolo is a Mac app for everything you save: screenshots, saved pages, images, and the downloads you meant to get to. It files them into your own folders, finds them by words, meaning, picture, or color, and lays them out on canvases. This is its MCP server. With it, Claude, Cursor, or any other MCP client can work with your library:
"Find the screenshots I saved last week."
"Look at this canvas and tell me what these have in common."
"Name and file everything in my Inbox."
This server needs the Studiolo app. It is a thin client of the app running on your own Mac, so it does nothing on its own. Studiolo is in a private alpha; join the waitlist to get it.
Tools
Tool | What it does |
| Searches the library by text, and also by meaning and by picture. |
| Returns the images themselves, so the assistant can actually look at them. |
| The most recently saved items, newest first. |
| The folders the library reads, with item counts. |
| Your canvases, by project, with how many items each holds. |
| What is on one canvas, in reading order, with each item's description. |
| Writes the name and note for one item, which makes it findable. |
| Renames a file on disk. |
| Files an item into a different folder. |
It cannot delete anything or change a canvas, and it will not move a file that is already on one.
Related MCP server: daguerre
Privacy
Everything stays on your Mac. The server talks only to Studiolo at localhost, and Studiolo sends nothing home. What your assistant does with what it sees is between you and your assistant.
Setup
Install Studiolo and keep it open while you use your assistant.
In Studiolo, open Settings › AI assistant. It gives the steps for your assistant:
Claude Desktop: double-click
studiolo.mcpb, which Settings shows you in Finder.Claude Code: one command to paste into Terminal.
Cursor and others: a block to paste into their MCP config file.
From this repo
npm install
npm run buildThen point your MCP client at node /path/to/studiolo-mcp/dist/index.js. Studiolo is expected at http://localhost:4444; set INSPO_CANVAS_URL if yours runs elsewhere.
Connected it?
Tell us. Studiolo counts nothing, so this is the only way we know it works for you. It is one field and asks for nothing about you.
License
MIT for this server. The Studiolo app is a separate product and is not open source.
Made by Mer So Studio.
Available Tools
9 toolscanvas_itemsA
What is on one canvas, in reading order, with each item's description.
Returns paths and metadata, not pictures — pass the paths to view_inspo to actually look at the arrangement.
Order is spatial: top to bottom, then left to right within a row. That matters, because a canvas is an arrangement and the order things were added in says nothing about how they now sit beside each other. Text and sticky notes on the canvas are returned too, in position, since they are often the reason a grouping exists.
Inputs: canvas (required), project.
describe_itemA
Write the name and note for one item in the archive.
This is the tool that earns the server its keep. Describing an archive by hand is the slow, valuable work that makes everything else findable — search by meaning has nothing to rank without it, and list_folders will tell you exactly which folders are thin.
Look at the image with view_inspo FIRST, then describe what is actually there. A guess from the filename is worse than no description: it will be searched, believed, and never checked.
The NAME is one line — what this is, in the words somebody would look for. It becomes the item's name everywhere in the app. The NOTE is the longer thought: why it was kept, what to steal from it. Either can be left out to keep what is already written; passing an empty string clears it.
Writes to the _manifest.md file beside the image, in plain Markdown, so the words outlive this app.
Inputs: path (required), name, note.
list_canvasesA
List the canvases in the person's Studiolo projects, with how many items each holds.
A canvas is a composition — things they deliberately put next to each other — which is a different thing from a folder, where items merely share a category. Reviewing a canvas is therefore a much more pointed question than reviewing the archive.
Canvases live in the canvas app's browser storage and are mirrored to disk on every change, so what is reported here is the state as of the last edit.
Inputs: no named parameters.
list_foldersA
List the folders the library reads, with their item counts and how many items in each carry a written label.
The archive may span more than one root folder on disk — each is listed with its own folders nested under it, and a folder is named "@/" so the two can be told apart. Passing just the folder name to search_inspo or recent_inspo matches that folder in any root.
The labeled count is the useful number: it says where the archive can be searched by meaning and where it can only be searched by filename. A folder showing "0 labeled" holds material that text search effectively cannot reach.
Inputs: no named parameters.
move_fileA
File an item into a different folder. This is the other half of describe_item: describing makes a pile findable, filing makes it tidy, and both are the slow work somebody drops things into Inbox to avoid doing by hand.
Use list_folders first to see what folders exist. The folder is created if it does not exist yet.
REFUSES for a file that is on a canvas, and the reason is worth knowing rather than working around: canvas elements store the path of the file they show, those elements live inside the app, and only the app can rewrite them. Studiolo's own move updates every canvas as it goes. A move made from out here would leave those canvases pointing at nothing, with no way for either side to notice. Move it in Studiolo instead.
REFUSES ALSO when it cannot find out whether the file is on a canvas. Missing information is not permission.
The filename and its extension are unchanged — use rename_file for the name. Nothing is overwritten: a collision gets a numbered suffix. Any description the file has travels with it.
Inputs: path (required), folder (required).
recent_inspoA
The most recently saved items, newest first.
Useful for "what did I collect this week", and for finding things clipped from the browser extension that have not been sorted or described yet.
Inputs: limit, folder.
rename_fileA
Rename the file on disk. This changes the filename, not the item's name — use describe_item for that. Most of the time describe_item is what you want: the name is what the app displays, and a camera filename is only worth changing when it is actively misleading.
REFUSES for a file that is on a canvas. Renaming one would leave every canvas element pointing at nothing; the app's own rename fixes those up as it goes, and only the app can, so do it there.
The extension is kept and cannot be changed — it describes the bytes, not the name.
Inputs: name (required), path (required).
search_inspoA
Search the person's local visual reference archive by text.
Matches across the filename and folder, a saved page's title, URL and note, and the _manifest.md label and note written for an image. Every space-separated term must match, so "cosmos paywall" narrows rather than widens.
Returns metadata only — use view_inspo afterwards to actually look at the results. Prefer searching before viewing: viewing is expensive and this is cheap.
Searches three ways at once, exactly as the app's own search box does, and says which is which: an exact match on the words; then anything the phrase reaches BY MEANING; then, if the person has switched it on, anything that looks right BY PICTURE — which is the only one that reaches an image nobody has ever described.
Supports the app's full search language: "quoted phrases" hold words together, and -negation leaves things out. Words match whole, so "red" does not match "centered".
Inputs: query (required), kind, limit, folder.
view_inspoA
Look at specific items from the archive. Takes paths from search_inspo or recent_inspo.
Returns the images themselves so they can actually be examined — composition, color, type, layout. This is the tool that makes the archive useful rather than merely listed.
Images come back downscaled to roughly 640px on the long edge, which is enough to read a layout and a palette. Ask for full resolution only when fine detail genuinely matters. Videos return their poster frame, since a still is all that can be shown.
Costly: each image is roughly 1-2k tokens, and at most 12 are returned per call. Search first and choose deliberately.
Inputs: paths (required), fullSize.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 9 tools
Each tool has a clearly distinct purpose: search vs. recent listing vs. viewing images, folder listing vs. canvas listing vs. canvas item inspection, and describing vs. renaming vs. moving files. The descriptions explicitly clarify boundaries, especially describe_item versus rename_file and move_file.
Most names use snake_case and a clear verb_noun pattern (search_inspo, view_inspo, list_folders, list_canvases, describe_item, rename_file, move_file). A couple deviate, such as recent_inspo and canvas_items, but they remain readable and domain-specific.
Nine tools is well-scoped for a visual reference archive: search, browse, inspect, describe, and organize. No tool feels redundant, and the surface is small enough to remain understandable without being thin.
The core archive workflow is covered: search, view, describe, rename, move, and inspect folders/canvases. There are minor gaps such as no explicit delete/remove tool and no canvas-editing operations, but the most important agent-facing workflows are present.
Maintenance
Related MCP Connectors
Read-only MCP server exposing a user ORANO library to their own AI agent.
Browse, search, upload, tag, transfer, and delete images in your Raster libraries over MCP.
Remote MCP server for full read/write access to a Zotero library
Agentic search over your Dewey document collections from any MCP-compatible client.
Related MCP Servers
AlicenseNot gradedqualityFmaintenanceEnables local agents to browse, import, organize, nondestructively edit, and export photos in a running desktop photo catalog, with revision-checked edits, previews, and batch operations over an authenticated loopback endpoint or bundled stdio adapter.MIT- AlicenseNot gradedqualityBmaintenanceEnables MCP clients to search, filter, and manage a private local photo library with CLIP-based text and similarity search, EXIF/GPS metadata, duplicate detection, albums, and folder indexing, returning ranked results and optional contact sheets.MIT
- AlicenseNot gradedqualityBmaintenanceProvides a local creator asset library MCP server that scans folders to catalog media files, batch-renames them, applies tags, tracks asset lineage between source and finished works, and offers keyword- and BGE-vector-based semantic search. It also lets other content-production MCP servers auto-register their generated outputs into the library for later retrieval and reuse.MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients to search and read a local library of saved prompts, skills, MCP servers, and design boards, fill prompt blanks, and retrieve board design briefs through read-only, offline tools, prompts, and resources.MIT