Beemm Vision
Drive Beemm Vision from Claude or Cursor: read projects, workflows and gallery, build node-based workflows, and generate images and videos with 98+ AI models (Nano Banana, Kling, Seedream, FLUX…).
Hosted MCP Server
npx add-mcp 'https://app.beemmvision.com/mcp'Installs into Claude Code, Codex, Cursor and more
Documentation
Claude · ChatGPT
The connector you add in a minute reads projects, workflows, models, balance, jobs and gallery; builds and runs workflows; generates images, alone or in batches of up to 12, and video; and edits an image (upscale, background removal, inpaint with a mask given by URL, expand). Spending is capped per account: 200 images and 50 clips a day, 20 generations per ten-minute window. Each batch entry counts, and each paid node of a run counts against the 50 clips; a run holds at most 30 nodes, 20 of them paid.
guardrails · caps, revocation
Claude (claude.ai and Desktop) and ChatGPT can build and generate once you approve the access screen. Claude Code, Cursor and any other MCP client connect read-only for now: projects, workflows, gallery, balance. It works on your personal projects. Spending tools take an idempotency key, so a call retried with the same key is billed once, and you can revoke access anytime in Réglages → Applications connectées (Settings → Connected apps).
Install it once, in the client you already use
Paste the address. Sign in when your client asks. That is the whole setup.
- 01Customize → Connectors → + Add → Add custom connector.
- 02Name it “Beemm Vision”, paste the server URL below, then Continue.
- 03Click Add and sign in with your Beemm Vision account when Claude asks (OAuth). Nothing to install.
https://app.beemmvision.com/mcp
Works on every Claude plan (one custom connector on Free). In a chat: + → Connectors → Beemm Vision.
Start from a prompt
Copy one, paste it into your agent. Each one works with the connector as it ships today.
works in Claude and ChatGPT
Generate one image, from the chat
Generate an image in my Beemm Vision project “<name>”: <subject>, in <style>. Tell me what it will cost, start it, and give me the job id.
works in Claude and ChatGPT
Review what I made this week
Show me my 8 latest Beemm Vision generations as a grid, then tell me which project each one belongs to and what it cost.
works in Claude and ChatGPT
Read a workflow like an engineer
List my Beemm Vision workflows in project “<name>”, open the most recently updated one, and explain node by node what it does and which models it calls.
works in Claude and ChatGPT
Where do my credits go?
Check my Beemm Vision balance, list my last 10 jobs and runs, and summarise what consumed credits, grouped by model.
works in Claude and ChatGPT
Four variants, one card
Plan four variants of “<subject>” on four different image models in one batch, show me the cost first, then generate them.
works in Claude and ChatGPT
Turn a still into a clip
Take the image I just generated and animate it into a short clip. Show me the cost first if Beemm Vision can price it before launch.
works in Claude and ChatGPT
Run a workflow, then finish the image
Duplicate my Beemm Vision workflow “<name>”, run the copy and wait for the result, then upscale the final image 2× with a credit cap you show me first.
Every tool your agent can call
Generated from the server itself. Through the connector, paid tools answer at once with a job or run id that the agent then follows, and a batch holds up to 12 images.
| Tool | What it does | Where | Credits |
|---|---|---|---|
| beemm_whoami | Reports which Beemm Vision account this MCP server is acting as, when that session expires, whether it carries an App Check attestation (required by most backend endpoints), and which backend it talks to. | connector | free |
| beemm_selftest | Runs the full Beemm Vision chain end to end in one call — credential, session expiry, which backend is configured and why, a real read against that backend, and whether any data is actually readable — and returns one line per check with the cause AND the remedy for anything that failed. | connector | free |
| beemm_list_projects | Lists the Beemm Vision projects you can access, most recently updated first, each labelled PERSONAL or ORGANISATION. | connector | free |
| beemm_list_workflows | Lists the workflows of one project, most recently updated first. | connector | free |
| beemm_get_workflow | Returns a workflow's nodes and edges, PROJECTED AND BOUNDED so it fits in context: per node its id, type, label, position and whether it is bypassed (generates nothing); per input its key, value type, whether it is fed by an edge or typed in, and a ~200-character preview. | connector | free |
| beemm_get_balance | Returns your PERSONAL credit balance. | connector | free |
| beemm_list_models | Lists the image and video models this account can actually run, with what each one supports: aspect ratios, resolutions, durations, whether it takes input images, whether it can generate audio. | connector | free |
| beemm_list_node_types | Lists every node type a workflow can hold, with what beemm_patch_workflow will accept for each: its category, its input handles (the toHandle an edge can reach) and output handles (the fromHandle an edge can leave from), its input keys with their default values, and its default size. | connector | free |
| beemm_plan_generationrenders a card | Shows the user a card with the exact image request — prompt, model, aspect ratio, resolution, reference images — and a button that starts it. | connector | free |
| beemm_plan_batchrenders a card | Shows the user ONE card listing up to 12 image requests — each with its prompt, model, aspect ratio, resolution and reference images — the total the backend would charge, and a single button that starts them all. | connector | free |
| beemm_generate_imagerenders a card | Starts an image generation and returns its jobId IMMEDIATELY — it does not wait. | connector | spends |
| beemm_generate_image_batchrenders a card | Starts up to 12 image generations in ONE call — typically the same prompt across different models, or several variations of one idea — and returns their job ids IMMEDIATELY: it does not wait. | connector | spends |
| beemm_generate_videorenders a card | Starts a video generation and returns its jobId IMMEDIATELY — it does not wait. | connector | spends |
| beemm_check_jobrenders a card | Follows ONE generation job, identified by a job id you already have: status, the credits the SERVER charged, and — once finished — the media URL. | connector | free |
| beemm_check_jobsrenders a card | Re-reads SEVERAL generation jobs of one batch in a single call and answers in the batch shape (one entry per job, in the order given, plus the total the server charged so far). | connector | free |
| beemm_list_jobs | Lists your most recent generation jobs and workflow runs, newest first, with their ids and statuses. | connector | free |
| beemm_list_galleryrenders a card | SHOWS the media an account already holds, as a single GRID of thumbnails, newest first, with their URLs. | connector | free |
| beemm_run_workflow | Executes an existing workflow on the server and returns immediately with a run id. | connector | spends |
| beemm_get_run | Reads a workflow run: its status, how many nodes are done, the credits the SERVER charged, and the output of each finished node (media URL, or text). | connector | free |
| beemm_create_workflow | Creates a NEW, EMPTY workflow — no nodes, no edges — ready for beemm_patch_workflow to build a graph in. | connector | writes |
| beemm_duplicate_workflow | Copies an existing workflow into a NEW project of its own — one workflow per project is the app's model, so the copy gets its own card on the Workflows page, with workflow id "default". | connector | writes |
| beemm_patch_workflow | Edits a saved workflow: add, update and remove nodes, connect and disconnect them. | connector | writes |
| beemm_upscale_image | Enlarges an image 2× or 4×, keeping the framing. | connector | spends |
| beemm_remove_background | Removes the background and returns a PNG with real transparency, at the source resolution. | connector | spends |
| beemm_inpaint_image | Changes ONLY the area a mask marks, and leaves the rest of the pixels alone. | connector | spends |
| beemm_expand_image | Keeps the picture at its exact size and paints NEW surroundings around it — the way to turn a portrait crop into a banner without stretching or cropping anything. | connector | spends |
| beemm_open_in_agent | Opens a workflow in YOUR OWN browser tab, signed in as the user, with the page's live tools armed — what the « Open in agent » button does in the web app, without the user clicking. | connector | writes |
Questions we get about the MCP
A server that exposes Beemm Vision to AI agents through the Model Context Protocol. Once connected, Claude or ChatGPT can list your projects, read, build and run your workflows, show your gallery, check what a job cost, generate images and video and edit an image; Cursor and Claude Code can read for now. The agent calls tools; Beemm Vision does the work, and in claude.ai answers with cards.
See also the model catalogue.
Keep reading
- Step-by-step guide: generate from Claude or ChatGPTSetup, first requests, limits and what to do when generation does not show.
- The models your agent can pickEvery image and video model, with what each supports.
- How credits workWhat each generation costs, and how plans renew.
- Ownership and training dataWhat you own, and what we never train on.
- The Beemm Vision API: run your workflows over HTTPFor a script or a backend rather than an AI assistant.