GenPresso MCP
Connect AI assistants like Claude and ChatGPT to GenPresso. Generate images and video right from the chat you already use — results land on your GenPresso canvases, ready to keep editing in the studio.
What is the MCP integration?
MCP (Model Context Protocol) is an open standard that lets AI assistants use external services as tools. Once you connect the GenPresso MCP server, your assistant can drive GenPresso's generation features from plain conversation.
- Generate by chatting — say "make an image of a girl walking down an alley" and the assistant calls GenPresso's tools.
- Results land on canvases — each generation becomes a node on a GenPresso canvas, so you can continue with upscaling, editing, and timeline work in the studio.
- No API keys — you sign in to your GenPresso account once in the browser when connecting. Credits are spent from that same account.
Connect from Claude
claude.ai (web) and Claude Desktop connect via custom connectors. Your plan must include the connectors feature.
- ① Open Claude Settings → Connectors.
- ② Click "Add custom connector", name it GenPresso, and paste the URL below.
- ③ A browser window opens — sign in with your GenPresso account and approve the connection.
- ④ When GenPresso appears under the tools icon, you're done — ask things like "use GenPresso to make an image of a sunset over the sea."
https://genpresso.ai/api/mcpIn Claude Code (terminal), one command adds it. The browser sign-in opens on the first tool call.
claude mcp add --transport http genpresso https://genpresso.ai/api/mcpConnect from ChatGPT
- ① In ChatGPT Settings → Connectors, enable Developer mode.
- ② Create a new connector and enter the same MCP server URL as above.
- ③ Choose OAuth as the authentication method and follow the prompts to sign in to GenPresso and approve.
- ④ Enable the connector in a conversation and ask for GenPresso tools.
Sign-in and security
Connections use standard OAuth 2.1. Adding the connector opens a GenPresso sign-in and consent screen in your browser; after you approve, the client stores an access token. You never create or paste an API key.
- Tokens are stored as hashes only — GenPresso's servers never keep the raw token.
- Access tokens last 1 hour and refresh tokens 30 days; the client renews them automatically.
- To disconnect, remove the connector in your client (Claude/ChatGPT). An expired refresh token requires signing in again.
Available tools
A connected assistant can use the tools below. You usually don't call them by name — describe the result you want and the assistant composes the calls.
| Tool | What it does |
|---|---|
| get_credit_balance | Check your current credit balance |
| list_models | List available image and video models |
| list_canvases | List your canvases (projects) |
| create_canvas | Create a new canvas |
| generate_image | Generate an image on a canvas — supports base/reference image inputs (asset_id) |
| generate_video | Generate a video on a canvas — supports start-frame/reference image inputs (asset_id) |
| check_generation | Check a job's status and result — image outputs include an inline thumbnail preview |
| jobs_wait | Wait server-side for jobs to finish (no polling loop) |
| list_assets | List generated assets — permanent asset_ids with download URLs |
| get_asset | Re-issue an asset's download URL, with an image preview |
Credits and limits
- Generation costs are identical to generating in the studio, and are deducted from the signed-in account's credits.
- If a generation fails, the deducted credits are refunded automatically.
- To prevent abuse, generation requests are limited to 30 per minute, plus a per-plan cap on concurrent jobs.
- Check your balance by asking for get_credit_balance in chat, or in the studio's top-right corner.
Results land on canvases
Everything generated over MCP is created as a node on a canvas. The assistant picks one of your existing canvases or creates a new one, and each response includes a link that opens that canvas directly.
- Generation is asynchronous — the assistant confirms completion via jobs_wait / check_generation (video can take several minutes).
- If the canvas is open, results appear automatically on completion; opened later, the results are already filled in.
- MCP-created nodes are ordinary nodes — the upscale button, @element references, and timeline placement all work as usual.
- Download URLs are served from the genpresso.ai domain — in egress-allowlisted environments (agent sandboxes, corporate networks), allowing just genpresso.ai is enough.
Troubleshooting
- Connection or auth errors (401): remove the connector in your client and add it again to re-authenticate.
- Tools not listed: restart the client or toggle the connector off and on.
- Generation seems slow: that can be normal — video models take minutes. Ask the assistant to check the status.
- Insufficient credits: top up on the studio's pricing/billing page and try again.
If you need a REST API to call from code, see the API docs (button in the header). For anything else, reach support via the footer.