--- name: use-sloosh description: Connect an agent to the Sloosh MCP server, sign in with OAuth, then run spaces and generate images and video within the workspace's daily limit. --- # MCP Connect Claude, ChatGPT, Cursor, VS Code or Codex to your Sloosh workspace over MCP to run spaces, generate and build canvases within a daily limit. Last updated: 2026-10-31 **Key facts** - The SpaceLab MCP server lets AI tools such as Claude, ChatGPT, Cursor, VS Code and Codex work in your Sloosh workspace: run spaces, generate images and video, manage assets and build canvases. - Until someone sets a daily limit in MCP settings, Sloosh MCP's limit is 50 credits a day and every paid run asks first; every connected AI tool shares the limit, it resets at midnight UTC and the agent can't raise it. - Every paid Sloosh MCP run starts with a quote that is good for 5 minutes, so the agent sees the price before anything is charged. - Sloosh MCP has three permissions: `spacelab:read` to read, `spacelab:author` to edit and publish drafts without spending, and `spacelab:execute` to spend credits. - Sloosh MCP is on every plan. Web page: https://sloosh.ai/docs/mcp ## What MCP is MCP (Model Context Protocol) is an open standard that lets an AI assistant use tools from other apps. The SpaceLab MCP server gives your assistant tools for your Sloosh workspace, so you can ask it to do the work in plain words. ## What your agent can do - Read your workspace, plan and credit balance. - Find assets in your library, import an image or video from a link, or upload a file. - Find a [space](https://sloosh.ai/docs/spaces.md), read its form, get a quote and run it. - Pick a [model](https://sloosh.ai/docs/models.md), get a quote and generate images or video. - Work in your [projects](https://sloosh.ai/docs/projects.md): create and open them, add folders, move generations and elements, add uploads and save a generation to a project or folder. - Build a canvas: create a draft, edit it, test it once and publish it as a space. Sloosh MCP is on every plan, and your starter credits work with it. [Pricing](https://sloosh.ai/pricing) has the details. ## Connect an AI tool Add the server address to your tool once. Then sign in with your Sloosh account in the browser. You never paste a password or key into the tool. **Server address** ```text https://mcp.sloosh.ai/mcp ``` ### Claude Code Run this command, then run `/mcp` and sign in. **Terminal** ```text claude mcp add --transport http sloosh https://mcp.sloosh.ai/mcp ``` ### Claude web and desktop Add Sloosh to Claude as a plugin or as a custom connector. The plugin also adds skills that guide Claude through product images, ad variants, spaces, product video and UGC storyboards. Its launch film skill renders a product launch film with Remotion and needs a coding agent that runs commands on your computer, such as Claude Code. To add the plugin, download it from the [MCP page](https://sloosh.ai/mcp) on sloosh.ai. In Claude, open Customize, then Plugins. Choose Add, then Upload plugin, and pick the file. Open the plugin's Connectors tab. Choose Add if Sloosh isn't added yet, then Connect, and sign in when the browser opens. On a Team or Enterprise plan, an Owner adds the connector for the organization first. To add a custom connector instead, open Customize, then Connectors. Choose Add, then Add custom connector. Name it Sloosh and paste the server address. Sign in when the browser opens. Custom connectors depend on your Claude plan. After you connect, open the connector's settings in Claude. Set the read-only tools to always allow, so reads stop asking for approval. Keep the other tools on ask, so Claude checks with you before it changes your work or spends credits. ### ChatGPT Install the [Sloosh plugin](https://chatgpt.com/plugins/plugin_asdk_app_6ae8d6d5804c8191ba0b9f9b60d1fae2) from ChatGPT's plugin directory and sign in when the browser opens. To connect a custom server instead, open Plugins on ChatGPT web, choose the plus button, then Add custom MCP server. Name it Sloosh, paste the server address, set authentication to OAuth, accept the risk warning and choose Create as a plugin. For a custom server, OpenAI gives full MCP, including tools that spend credits, to ChatGPT Business and Enterprise/Edu workspaces, where an admin or owner adds the server. On Pro, a custom server gets only read and fetch tools, so it can browse your workspace but can't generate or run spaces. To use Sloosh from any ChatGPT plan, connect it in [Codex](https://sloosh.ai/docs/mcp.md#codex) instead. **Server address** ```text https://mcp.sloosh.ai/mcp ``` ### Cursor Add this to `.cursor/mcp.json` in your project, or to your global MCP file. Connect the server in Cursor and sign in. **.cursor/mcp.json** ```json { "mcpServers": { "sloosh": { "url": "https://mcp.sloosh.ai/mcp" } } } ``` ### VS Code Add this to `.vscode/mcp.json`. Run MCP: List Servers, start `sloosh` and sign in. **.vscode/mcp.json** ```json { "servers": { "sloosh": { "type": "http", "url": "https://mcp.sloosh.ai/mcp" } } } ``` ### Codex Run these commands. The second one opens the browser to sign in. **Terminal** ```text codex mcp add sloosh --url https://mcp.sloosh.ai/mcp codex mcp login sloosh --scopes spacelab:read,spacelab:author,spacelab:execute,user:org:read,offline_access ``` ### Other tools Any tool that connects to remote MCP servers over HTTP, with sign-in, can use the same address. For a tool that only starts local servers, use the `mcp-remote` bridge. It opens the browser to sign in. **MCP config** ```json { "mcpServers": { "sloosh": { "command": "npx", "args": [ "-y", "mcp-remote@0.14.3", "https://mcp.sloosh.ai/mcp", "--transport", "http-only" ] } } } ``` ## Set a daily limit Until you set one, the limit is 50 credits a day and every paid run asks first. After you connect, switch to that workspace, open [MCP settings](https://sloosh.ai/mcp#agent-settings) and set the limit. ## Starter prompts Once your tool is connected, paste one of these. Every paid run starts with a quote. - Make a studio photo of my water bottle on a clean white background. - Create three 9:16 ad variants for my summer sale on this sneaker. - Find a space in my workspace that makes product ad grids and tell me what a run costs. - Turn my latest product image into a short video clip. - Make three storyboard frames for the Episode 1 folder of my Night Shift project. - Move my last three mug images into the Lookbook folder of my Summer Launch project. ## Choose your workspace When you sign in, you pick the workspace the connection works in. The connection stays in that workspace. To work in another one, connect again and pick it. Ask your agent to call `get_connection` first. It shows the workspace, the permissions you granted and how much of today's limit is left. ## Permissions You grant three permissions when you sign in: - `spacelab:read` lets the agent read your workspace, assets, spaces, models, runs and credit balance. - `spacelab:author` lets it read, create and edit draft canvases, price a test run and publish them as spaces. It never lets the agent spend credits. - `spacelab:execute` lets it spend credits: run spaces, generate and test drafts. It also lets the agent import and upload assets, and change your projects. A connection keeps the permissions it had when you signed in. If you connected before canvas editing was available, connect again to grant `spacelab:author`. The Codex command also asks for `user:org:read`, so the connection can see the workspace you picked, and `offline_access`, so you stay signed in. In Claude, the connector's settings group its tools into read-only tools and tools that make changes. Set the read-only tools to always allow, so reads stop asking for approval. Keep the other tools on ask, so Claude checks with you before it changes your work or spends credits. ## Credits and limits Agent runs spend your workspace credits, the same as runs in the app. [Plans and credits](https://sloosh.ai/docs/plans-and-credits.md) explains how credits work. MCP adds these guards: - Until someone sets a daily limit in [MCP settings](https://sloosh.ai/mcp#agent-settings), the limit is 50 credits a day and every paid run asks first. Every AI tool connected to that workspace shares it, and the agent can't raise it. - The limit resets at midnight UTC. Runs still in progress count against it until they finish, and reconnecting doesn't reset it. - Every paid run starts with a quote. The agent sees the price before anything is charged, and a quote is good for 5 minutes. - In MCP settings, Ask before spending more than sets when the agent must confirm. At 0, every paid run needs a quote and an explicit confirmation first. Above 0, the agent can run anything quoted at that amount or less without asking. - Each run has a key. When the agent retries after a lost response, it reuses the key, so you're charged once. - If you cancel a run, finished outputs stay charged. Unused credits come back when the run ends. ## Canvas editing - The agent edits drafts only. A published space doesn't change until someone publishes again. - Every agent edit saves a version, labeled with your AI tool's name, such as "Claude Code edit". You can restore any version from the canvas's version history. - A test run of a draft may fan out to at most 8 paid node runs in total. One run can make several images, so 8 runs can make more than 8 images. A test run can't fan out by a count only known while it runs, such as an iterator fed by a text model. It needs a quote and `spacelab:execute`, like any paid run. - Publishing is its own step. The agent previews it and shows you the summary first, then publishes. Only a workspace admin can publish. - The agent publishes to your workspace only. Publishing replaces the canvas's earlier published space, if it has one. If someone made that space public, the agent can't replace it; it stays public and unchanged. - A canvas that Goal Mode (an AI builder some workspaces have in Spacelab) is working on stays locked. The agent can read it but can't edit, run or publish it. ## Disconnect a tool [MCP settings](https://sloosh.ai/mcp#connected-agents) lists your connections. Disconnect one to stop new calls right away. Runs already under way still finish and are charged. To use it again, choose Allow reconnection in MCP settings, then remove the server from your tool, add it back and sign in. Today's usage doesn't reset. ## Tools Your AI tool shows the full list with each tool's inputs. The tools are grouped here by what they do. ### Connection and guides - `get_connection`: the workspace, permissions and today's remaining limit. - `get_guide`: read one guide by its id. With no id, it lists the built-in guides for setup, spending and each workflow. ### Account - `get_account`: your workspace, plan, MCP access and credit balance. ### Assets - `list_assets`: list library or generated assets. Pass a name to search them, or a project (and a folder, for generations) to list what's in it. - `get_asset`: read one asset and its link. - `import_asset`: add an image or video from an https link, up to 64 MiB. - `initiate_asset_upload`: start uploading a file, up to 64 MiB. - `complete_asset_upload`: check the uploaded file and add it to the library. ### Projects - `list_projects`: the projects you can see, your Private ones and your workspace's Team ones, each with its folders. Pass `includeArchived` to add archived ones. - `get_project`: open one project: its brief, who can see it, its defaults and its folders with their counts. - `create_project`: create a project. It starts Private. - `update_project`: change a project's name, brief or defaults. - `create_project_folder`, `rename_project_folder`, `delete_project_folder`: manage a project's folders. Deleting a folder moves its items to the project's root. - `move_items`: move generations and elements into a project, up to 50 at a time. Generations can also go into one of its folders; elements go into the project itself. A run moves as one. - `add_uploads_to_project`: add files from your library to a project. They stay in the library too. - `make_project_team`: make a Private project you own a Team project. It can't be undone, so the agent can't do it on its own. If your AI tool can ask you directly, it asks you to approve sharing the project. Otherwise the agent gets a confirmation step that lasts 5 minutes and has to show you the project's name and get your yes first. If you decline, nothing changes. Archiving, restoring and deleting a project are done in the app, in the project's Settings. ### Generation - `list_models`: the image and video models you can use, with their options. - `enhance_prompt`: rewrite a short prompt into a detailed one for the chosen model, free, so you can read it before you quote. - `quote_generation`: price an image, a video or a video edit. Pass a project, and a folder if you like, to save the results there. - `generate`: run a quoted generation. - `get_generation`: read status, outputs and the final charge. - `cancel_generation`: cancel a generation. - `reconcile_generation`: recover a generation after a lost response. ### Spaces - `list_spaces`: find published spaces in your workspace. - `get_space`: read one space. - `get_space_input_schema`: read a space's form. - `quote_space`: check the inputs and price a run. - `run_space`: run a quoted space. - `get_space_run`: read a run's progress, outputs and any step waiting for approval. - `get_space_run_outputs`: read a run's outputs so far. - `approve_space_checkpoint`: approve a step that waits for approval. - `reject_space_checkpoint`: reject a step that waits for approval. - `cancel_space_run`: cancel a run. - `reconcile_space_run`: recover a run after a lost response. ### Canvas authoring - `list_node_kinds`: the node types a canvas can use. - `search_examples`: find example canvases similar to a goal. - `list_templates`: list the starter templates from the canvas template gallery. - `get_canvas`: read a canvas and whether it's locked. - `validate_canvas`: check what still blocks a run. - `list_canvas_versions`: list saved versions, newest first. - `create_draft`: create a draft canvas, empty or started from a starter template or an example. - `apply_canvas_operations`: apply a set of edits, all or none. - `restore_canvas_version`: restore a saved version. - `quote_draft`: price one test run of a draft. - `run_draft`: run a quoted draft once. - `get_draft_run`: read a test run's status, outputs and charge. - `preview_publish`: check a canvas can be published and get a summary. - `publish`: publish the previewed canvas as a workspace space. ## Frequently asked questions ### How do I connect Claude to Sloosh? In Claude web or desktop, download the Sloosh plugin from the [MCP page](https://sloosh.ai/mcp) on sloosh.ai, upload it in Customize, then Plugins, and open its Connectors tab. Choose Add if Sloosh isn't added yet, then Connect. Or open Customize, then Connectors, choose Add, then Add custom connector, name it Sloosh, paste the server address and sign in when the browser opens. In Claude Code, run `claude mcp add --transport http sloosh` with the address, then `/mcp` and sign in. You never paste a password or key. ### Can an AI agent spend my Sloosh credits? Only within the daily limit (50 credits a day until someone sets one in Sloosh MCP settings), and only after you grant `spacelab:execute`. Every paid run needs a quote first, and Ask before spending more than sets when the agent must confirm: at 0, every paid run needs an explicit confirmation. Agent runs spend workspace credits like runs in the app. ### Can an AI agent change my published Sloosh space? No. A Sloosh MCP agent edits drafts only, and a published space doesn't change until someone publishes again. The agent previews a publish and shows you the summary first, only a workspace admin can publish, and the agent can't replace a space someone made public. ### Can an AI agent use my Sloosh projects? Yes. A Sloosh MCP agent sees the projects you can see, your Private ones and your workspace's Team ones, with their folders. It can generate into a project, create projects and folders, move items and add uploads, and it asks you to confirm before it makes a project Team. Archiving and deleting a project are done in the Sloosh app. ### Can I undo an AI agent's changes to a Sloosh canvas? Yes. Every Sloosh MCP agent edit saves a version, labelled with your AI tool's name, such as Claude Code edit. You can restore any version from the canvas's version history. ### How do I connect ChatGPT to Sloosh? On ChatGPT web, open Plugins, choose the plus button, then Add custom MCP server. Name it Sloosh, paste the server address, set authentication to OAuth, choose Create as a plugin and sign in when the browser opens. Full MCP, which Sloosh needs to generate and run spaces, is on ChatGPT Business and Enterprise/Edu, where an admin or owner adds it; on Pro, ChatGPT allows read and fetch tools only. ### How do I connect Cursor or Codex to Sloosh? In Cursor, add the Sloosh server address under `mcpServers` in `.cursor/mcp.json`, connect the server and sign in. In Codex, run `codex mcp add sloosh --url` with the address, then `codex mcp login sloosh` with the scopes shown on this page, which opens the browser to sign in. ### How do I disconnect an AI tool from Sloosh? Open [MCP settings](https://sloosh.ai/mcp#connected-agents) and disconnect the connection to stop new calls right away. Runs already under way still finish and are charged. To use the tool again, choose Allow reconnection, remove and re-add the server in your tool and sign in; today's usage doesn't reset. ### I added the Sloosh plugin to Claude. Why can't Claude use Sloosh? Adding a plugin doesn't sign you in. In Claude, open Customize, then Plugins, open the Sloosh plugin and select its Connectors tab. Choose Add if Sloosh isn't added yet, then Connect, and sign in. On a Team or Enterprise plan, an Owner adds the connector for the organization first. ### Why does ChatGPT show old Sloosh tools? ChatGPT keeps a snapshot of the tool list it read when the Sloosh plugin was created. Changes reach your chats only after a workspace admin reviews and publishes an update to the plugin. Then start a new chat. ### My AI tool lost access to Sloosh. How do I reconnect? Open [MCP settings](https://sloosh.ai/mcp#connected-agents). If the connection was disconnected, choose Allow reconnection. Then remove Sloosh from your tool, add it back and sign in. Today's usage doesn't reset. ### Why won't my agent spend credits? There are three causes. The daily limit is 0: open [MCP settings](https://sloosh.ai/mcp#agent-settings) and raise it. Today's limit is used up: it resets at midnight UTC, or you can raise it in MCP settings. The agent lacks `spacelab:execute`: reconnect the tool and approve every permission. Until then the agent can read and build drafts, but it can't spend credits.