# Generalberrys Berrys Canvas MCP server

Source: https://glimind.com/server/mcp-registry%2Fio.github.generalberrys%2Fberrys-canvas
Registry id: `mcp-registry/io.github.generalberrys/berrys-canvas`
Status: live (checked 2026-11-22T19:08:43.021Z)

## Measurements (30 days)

- Uptime: 99.88%
- Checks: 862
- Outages (90 days): 1

## Install

```json
{
  "mcpServers": {
    "generalberrys-berrys-canvas-mcp-server": {
      "type": "http",
      "url": "https://app.berrys.io/api/mcp"
    }
  }
}
```

Claude Code:

```sh
claude mcp add --transport http generalberrys-berrys-canvas-mcp-server https://app.berrys.io/api/mcp
```

## Tools (70)

- `playground_actions_discover`: Discover implemented media editing and canvas actions. No project needed to browse; include projectId to check its permissions. By default, omit unavailable act
- `playground_approval_decide`: After showing the exact prompts, references and total credits in chat and receiving the user’s explicit confirmation, approve or reject one direct generation re
- `playground_approvals_list`: Read pending proposals, exact scopes, expiry, and price ceilings for the current user and project.
- `playground_asset_import`: Add one image (PNG, JPEG, WebP, ≤8 MB), video (MP4, ≤16 MB) or audio file (MP3 or WAV, ≤12 MB) to the canvas as a file node that can feed image and video genera
- `playground_brand_get`: Read the connected account’s selected brand and saved Brand DNA. Does not change the account’s active brand or apply the rules to a generation. Returns the sele
- `playground_brands_list`: List brands owned by the connected account. Use when the user wants to choose existing brand rules. Does not select an active brand, open a project or spend cre
- `playground_brief_start`: Start the Berrys agent on a brief and return at once with runId and canvasUrl; it never waits. Follow it with playground_run_events (afterSeq, optionally waitSe
- `playground_canvas_add_text`: Add an editable text node to a Playground workflow. Supply the latest canvas revision; stale edits are rejected. Returns the text node ID, updated revision and 
- `playground_canvas_apply_operations`: Apply one atomic, revision-checked, idempotent batch of non-paid canvas edits: create_node (text, image, video, video-stitch, comment, image-compare, output-gal
- `playground_canvas_arrange`: Arrange loose nodes as a block, or re-lay out a group’s children, as one revision-checked, idempotent, non-paid revision (the tidy operation of playground_canva
- `playground_canvas_open`: Return the authorized project URL focused on selected node IDs. If your host has an in-app browser, open it beside chat and reuse that project tab. Otherwise sh
- `playground_canvas_read`: Read the current node graph and revision for a Berrys Canvas Playground workflow. Returns the graph, current revision and Playground URL with a canvas preview.
- `playground_canvas_snapshot_restore`: Replace the whole canvas with a snapshot, as one new revision. The current canvas is snapshotted first, so the restore can itself be undone. Discards every chan
- `playground_canvas_snapshots_list`: List recent whole-canvas snapshots taken before agent runs, MCP writes, reverts and restores. Read-only. Returns snapshot IDs, revisions, reasons and node count
- `playground_canvas_update_node_layout`: Update only the common title and/or canvas position of any user-facing Playground node. This is non-paid, revision-checked, idempotent, and returns the previous
- `playground_canvas_update_text`: Replace the editable text content of an existing text node. Requires project edit access, the node ID and the current expectedRevision from canvas_read. Reuse i
- `playground_capabilities_discover`: Discover what you can make with Berrys without selecting or reading a project. Omit projectId for a general overview. For a chosen project, include projectId to
- `playground_connected_source_read`: Execute one allowlisted read with the exact connected account ID and discovered tool version. Returns bounded source data and a receipt; no external write or ex
- `playground_connected_source_schema`: Get the current exact input schema and tool version for one selected account and allowlisted read operation. Returns the operation version, input schema and pag
- `playground_connected_sources_list`: List active Drive, Shopify, Meta Ads, and Gmail accounts and read-only operation names. Uses the signed-in user’s connected accounts and project access. Returns
- `playground_content_prepare`: The easiest way to start making content. Give a short brief and one entry per output you need; intendedUse is optional. You write each image or video prompt you
- `playground_context_current`: Use only when the user asks to work on their open project or selection. Return lightweight context from an authorized editor active in the last 2 minutes. Never
- `playground_drive_image_import`: Copy one explicitly selected PNG/JPEG/WebP image (up to 8 MB) from your connected Google Drive account into Berrys reference storage, where generation providers
- `playground_external_link_inspect`: Fetch bounded metadata for an explicitly supplied HTTPS social-platform link. This does not watch its video or audio. Requires a creative-task grant because it 
- `playground_film_assembly_retry`: After two failed assembly attempts, validate the ordered Playground clips and queue another bounded FFmpeg assembly attempt. This does not submit a paid media g
- `playground_generation_confirm`: After explicit user confirmation of exact inputs and total credits, submit all selected direct-generation approvals from one run in one call. Pass confirmation.
- `playground_image_quote`: Read the current credit cost for one image and up to three image references. Does not submit a job. Returns model pricing and the quoted image settings.
- `playground_media_compare`: Compare 2–4 images or 2–4 videos side by side on the canvas. Two images fill a Compare (image-compare) node's A/B slots; three or four images feed a Picks (outp
- `playground_media_compose`: Stitch 2–12 finished video nodes, in the given order, into one film. Creates an "Assemble film" (video-stitch) node wired from the clips and queues the existing
- `playground_media_generate`: Batch-generate 1–20 images, videos or audio tracks in one request. Each image/video item becomes a new node, with its prompt Text node and wired referenceNodeId
- `playground_media_job_cancel`: Cancel one queued agent image or video job and release its credit hold. A processing provider job cannot be cancelled by this tool. Requires project edit access
- `playground_media_job_get`: Reconcile and read one media dispatch receipt by ID. Completed images include a native image preview, original media link and Playground URL; display the native
- `playground_media_jobs_list`: List recent agent image and video dispatch receipts for an authorized Playground project. Returns media job receipts and their current statuses.
- `playground_media_retry_quote`: Read fresh receipts for one run and quote replacements only for confirmed failed items, including individual List outputs. Completed, pending, active, uncertain
- `playground_media_upload_begin`: Create a short-lived signed URL for one image, video, or audio reference. Upload bytes to signedUrl with the declared Content-Type, then call playground_media_u
- `playground_media_upload_complete`: Verify the signed upload by server-owned path, actual byte size and media signature, then add one editable file node at the current canvas revision. Repeating t
- `playground_memory_list`: Read approved project-shared memory and this user’s private memory. Requires project view access. Returns the project’s available memories.
- `playground_memory_proposal_review`: Approve or reject one proposed memory note for this project, as the user would in Berrys. Approved notes are read by the Berrys agent on later turns; rejected o
- `playground_memory_propose`: Remember a user decision or verified project fact, with optional source node citations. It takes effect immediately: the Playground agent reads remembered notes
- `playground_memory_search`: Search approved project-shared memory and this user’s private memory by words in their content, with offset/limit pagination. Requires the memory:read grant. Re
- `playground_model_resolve`: Read-only. Resolve user shorthand such as Seedream 5.0, Nano Banana 2 or Seedance 2 Mini against enabled models. Case, punctuation and trailing .0 are normalize
- `playground_models_list`: Discover enabled image and/or video models with aspect ratios, durations, reference requirements, whether the agent can run them, and price: baseCredits (one im
- `playground_node_get`: Read one node by ID with its incoming and outgoing connections, its group children, and the current canvas revision. Cheaper than reading the whole graph.
- `playground_node_run`: Run the generation already configured on 1–20 existing image, video or audio nodes. Image: prompt (its Text node), modelId, aspectRatio, wired references. Video
- `playground_node_types_list`: Discover every Playground node type, what it consumes and produces, and which agent actions are currently supported. Use this before planning graph changes.
- `playground_nodes_search`: Search node IDs, types, titles, text, and prompts. Returns bounded summaries with pagination.
- `playground_output_feedback_list`: Read this user’s accept/change verdicts for media still attached to the Playground project. Requires project view access. Returns recorded output feedback for t
- `playground_project_create`: Create a new, empty Playground project owned by the signed-in user. For a fresh generation request with no named, linked or conversation-established project, cr
- `playground_project_skill_propose`: Submit instructional Markdown for human review in Playground. The proposal is inactive and grants no tools, connections, or billing rights. Returns skill and pr
- `playground_projects_list`: List Playground workflows that this OAuth client may read for the current user. Returns accessible project IDs, names and effects, with a truncation indicator.
- `playground_run_cancel`: Stop further agent effects for a run. Provider jobs already submitted may continue and retain their receipts and charges.
- `playground_run_delivery`: Read a run’s final response and persisted canvas or media receipts, with native image previews (up to four) and original media links. Display the native image c
- `playground_run_diagnose`: One report for a run: verdict (succeeded, failed, waiting_approval, running, cancelled), the failed step, the last good step, failed media jobs with errors, and
- `playground_run_events`: Return only what changed in a run since afterSeq: compact events, a status line, pending approvals, the final response once it ends, nextAfterSeq, and retry_aft
- `playground_run_revert`: Undo every canvas change one agent run made, newest first, as one new revision. If someone has edited those nodes since, nothing changes and the conflicting nod
- `playground_run_steer`: Add a note to a production plan that is running: it applies to steps that have not started and later outputs are reviewed against it. Steps already generating a
- `playground_skill_load`: Load exact versioned instructions. Built-in skills need no projectId. Private project skills require projectId and scope: project, with current read access. Doe
- `playground_skills_discover`: Explore built-in playbooks without opening or creating a project. Search by query/category and follow nextOffset. Include projectId only to include that project
- `playground_storyboard_prepare`: Create ready-to-generate video shot nodes connected to text nodes inside a selected text group. This only prepares nodes; it does not submit paid video jobs. Re
- `playground_video_analyze`: Read-only facts about one video node: its clip URL, model and settings, frame nodes already taken from it, objective FFmpeg measurements (duration, black segmen
- `playground_video_batch_quote`: Read the exact prepared text-to-video shot inputs and total credit ceiling in the requested order. This does not submit or approve jobs. Returns per-node video 
- `playground_video_extract_frames`: Take the first and/or last frame of a finished clip and place each on the canvas as a Frame Grab node wired from the clip, holding a real stored image. A frame 
- `playground_video_quote`: Quote a supported text-to-video model for one prepared clip. This is read only and does not submit a job. Returns model pricing and the quoted video settings.
- `playground_video_trim`: Cut a finished video node to [startSeconds, endSeconds] on the server. Creates a "Clip" (video-result) node wired from the source, as the in-app Clip tool does,
- `playground_web_page_read`: Fetch a public HTTPS page (product page, article, landing page, or social post) with private-network, size and type checks, and return its title, description, c
- `playground_workflow_inputs_set`: Save text/List inputs, prompts, formats, durations, audio settings and image reference choices from workflow_inspect. Free, atomic, revision checked and idempot
- `playground_workflow_inspect`: Read typed source input contracts, generation dependencies, completed outputs, pending/active work and blockers. Returns the next ready stage without spending. 
- `playground_workflow_quote`: Quote only the next ready empty steps from workflow_inspect, up to 20 outputs. Does not edit the canvas, prepare approval or submit generation. Safe for the cha
- `playground_workflow_run`: Run a saved workflow through its ready image/video/audio stages. quote:true prices up to 20 ready nodes including connected List batches. Pass the returned stag
- `playground_workflow_search`: Find accessible saved project workflows by name. Returns workflow IDs for inspection and staged execution. This does not search public templates.

Measured by Glimind (https://glimind.com): MCP handshake + tool list only, no tool is ever called. CC BY 4.0.
