OpenModels

Connect OpenModels to ChatGPT and Claude with MCP

Paste one URL into Claude, ChatGPT, Claude Code or another MCP app, sign in with your OpenModels account and press Connect. The app registers itself, so there is no key, client ID or configuration file to copy. Your assistant can then find OpenModels models, quote a job, submit it with your approval and retrieve the finished media from your library.

In short

Who can run this: Connecting an app is part of the Pro and Ultra plans. Generations spend that account’s own credits and follow its model access; the assistant’s own plan rules also apply.

Which plan: Connecting an app is part of the Pro and Ultra plans (see plans). Connecting costs nothing extra; each generation spends your own OpenModels credits after you approve its quote. Paying for Claude or ChatGPT does not add OpenModels credits, and an OpenModels plan does not include either assistant’s subscription.

What the OpenModels MCP connection does

MCP, the Model Context Protocol, lets an assistant call tools exposed by OpenModels. You can ask it to list image or video models, inspect their accepted inputs, find assets in your library and follow a generation job through to its result. OpenModels handles the generation and storage. The assistant provides the conversation that directs those tools.

The connection acts as the OpenModels account you signed in with. This matters when you work in several accounts: the job and resulting asset belong to that account, with its access and credit rules. A generation requested through an external assistant therefore appears alongside your other OpenModels work. A message saying “submitted” means a job was accepted; it does not mean the media is finished.

Connect any app with one URL

  1. Copy your MCP URL. Open Connections in the account menu and copy Your MCP URL, https://openmodels.vip/mcp. It is the same for every account and every app: it identifies the OpenModels server, not you, so it is safe to paste anywhere.
  2. Add it to your app. In Connections, pick your app. Claude, Cursor and VS Code have an Add button that fills everything in; for the others, paste the URL where the app adds a remote MCP server or custom connector (menus in the table below). The app registers itself with OpenModels, so leave any client ID or client secret fields empty.
  3. Sign in and press Connect. The app opens an OpenModels page. Sign in with the account that should own the jobs, check who is asking (see below), and press Connect. Untick Create and edit files if the app should only look. You are sent straight back to the app.
  4. Run a read-only check. Ask: “What is my OpenModels credit balance?” Confirm that a tool actually ran and the numbers match your account. A conversational answer without an OpenModels tool result does not establish that the account is connected.
Menu names at this guide’s last review. Apps rename their settings; look for connectors, apps or MCP servers.
AppWhere to add the URLThen
Claude (web, desktop and mobile)Add to Claude in Connections, or Customize → Connectors → Add custom connectorPress Add, then Connect, and sign in to OpenModels.
ChatGPTSettings → Apps → Developer mode, then Create appPaste the URL, keep OAuth sign-in, then Connect.
Claude Codeclaude mcp add --transport http --scope user openmodels https://openmodels.vip/mcpRun /mcp, select openmodels and choose Authenticate.
Cursor and VS CodeAdd to Cursor or Add to VS Code in ConnectionsInstall, then sign in to OpenModels when the app asks.
Gemini CLI and other MCP appsAdd a remote (Streamable HTTP) MCP serverUse the same URL; the app opens the OpenModels sign-in page.
Codex and clients without a sign-in windowAdvanced: account keys, in ConnectionsCreate a key and use the configuration Connections shows.

Connect Claude

Press Add to Claude in Connections. Claude opens its Add custom connector dialog with the name and URL filled in; press Add, then Connect. Claude opens the OpenModels sign-in page, and after you press Connect there it returns you to Claude with the connector enabled. Without the button: Customize → Connectors → Add custom connector, name it OpenModels and paste https://openmodels.vip/mcp. A connector added on the web also appears in Claude Desktop and on mobile. Claude Free allows one custom connector; on Team and Enterprise an owner adds it for the organization first, then each member presses Connect. See the official Claude custom connector guide.

Connect ChatGPT

In ChatGPT, open Settings → Apps, turn on Developer mode, then choose Create app. Paste the MCP URL, keep OAuth as the authentication method and create the app. ChatGPT opens the OpenModels sign-in page; press Connect and you return to ChatGPT. Developer mode depends on your ChatGPT plan and workspace permissions. This creates a personal app for your account, not a listing in a public directory. See OpenAI’s connection guide.

Connect Claude Code

Run this once in a terminal: claude mcp add --transport http --scope user openmodels https://openmodels.vip/mcp. --scope user makes it available in every project. Start Claude Code, run /mcp, select openmodels and choose Authenticate. Your browser opens the OpenModels sign-in page; after you press Connect it returns to a page served by Claude Code on your computer, and Claude Code keeps the sign-in for you. There is no key to paste and no environment variable to set. See the official Claude Code MCP guide.

Check the app before you press Connect

Any app can call itself anything, so the consent page names the app by where OpenModels would send you back. Claude and ChatGPT, returning to their own sites, show their name with a Verified mark. Every other app shows that address instead, such as example.com wants to connect, An app on this computer for terminal tools such as Claude Code, or The cursor app, with the name it chose underneath and an Unverified app badge. Unverified is normal for Claude Code, Cursor and VS Code; it means OpenModels cannot confirm the name. If a page claims to be Claude but is not marked Verified, or you did not start the connection yourself, press Cancel.

Connecting grants two permissions: See your files, prices and credits, and Create and edit files, spending your credits after you approve each price. Untick the second one to connect an app that can only look. To let it create later, disconnect it (on this page or in the app) and connect again with the box ticked. Connecting itself does not start a generation or spend credits, and every paid run shows its price and wallet in the app before you say yes.

See and disconnect connected apps

Connections lists every app connected to your account, with the address it signs in through, whether it is verified, what it may do (See, or See · Create & spend) and when it was last used. Choose Disconnect to revoke an app; it stops working with your account until you connect it again from the app. Signing out of OpenModels in your browser does not disconnect apps, because each app holds its own sign-in.

Advanced: account keys for clients without sign-in

Some clients cannot open a browser sign-in, such as Codex configured through ~/.codex/config.toml, scripts and CI jobs. For those, open Advanced: account keys in Connections, choose the client, name the key and select Create key. The complete key is shown only once; store it in the client’s credential storage or a password manager before dismissing it. Connections then shows the configuration for Codex, Claude Code or a generic client. Each example references the OPENMODELS_MCP_KEY environment variable rather than containing the secret. See the official Codex MCP guide.

Codex TOML settingValue
Section[mcp_servers.openmodels]
url"https://openmodels.vip/mcp"
bearer_token_env_var"OPENMODELS_MCP_KEY"

Revoke a key you no longer use from the same list. If you lose a key, create a replacement, update the client and revoke the old entry. You can have up to ten active keys; revoke an unused one if creation reports that limit.

Try a quoted image job, then retrieve the result

A useful first workflow has three distinct parts: inspect the model, quote the exact request and submit it once. Tell the assistant your output shape, reference asset and spending limit. For example: “Find my latest product photograph. Quote one portrait image using GPT Image 2, preserving the product, with a clean studio background. Show the account credit cost before submitting.”

The assistant can use list_models and get_model_schema to discover current options instead of guessing a model ID or accepted dimensions. A spending tool’s dry_run request is a quote, not a finished generation. Review the selected model, reference and settings; then ask the assistant to submit that quoted job. If you change the duration, model or other cost-relevant settings, request a fresh quote.

After submission, keep the returned job_id. Ask: “Check this job’s status and show the saved result when it finishes.” The assistant should call get_job_status with that ID, then return the actual media link. Repeating the generation request creates another job; checking status follows the original. A queued job may need time on shared capacity, so rapid repeated status calls do not make it render faster.

What it costs: MCP access does not make generation free. Model choice, settings, account allowances and the current quote determine the cost. Read how OpenModels credits work for the wallet rules.

Reuse media from your OpenModels library

Ask the assistant to use list_assets to find candidates and get_asset to inspect the selected item. Give a descriptive cue such as “the blue bottle studio photograph” and ask it to show the candidates before choosing if several match. Once selected, preserve the asset ID throughout the workflow so a later prompt refers to the same image.

To work from a video on YouTube, TikTok, Instagram, Facebook, X, Vimeo, Dailymotion, Twitch, Reddit, Streamable, Rumble or Bilibili, paste its link and ask the assistant to import it. It uses import_video_url, which downloads the video into your library as a background job; the finished job gives the new video’s asset ID for editing, captions or a start frame. Importing costs no credits, the file counts toward your storage, and only public https links are accepted.

For video, describe one intended shot and ask the assistant to inspect the model’s required inputs. An image-to-video model, an audio-driven talking model and a video-editing tool accept different material. Prepare individual speaker images for an interview and create separate shots when you want cuts between people. A storyboard sheet used as the actual image reference can appear inside the generated footage.

Troubleshoot connection and job errors

What you seeWhat to check
The sign-in page says the link is invalid or expiredStart again from the app you are connecting. Old sign-in links expire; the app creates a fresh one each time.
The consent page shows an address you do not recognize, marked UnverifiedPress Cancel. The app chooses its own name; the heading shows where OpenModels would actually send you.
Sign-in failed in the appRemove OpenModels from the app and add it again, then sign in once more.
The assistant answers without using OpenModelsStart a new chat and say “use OpenModels”. Check the connector is switched on for that chat.
401 or asked to sign in againReconnect from the app (in Claude Code, run /mcp and choose Authenticate). For an account key, check the environment variable and whether the key was revoked.
403, “part of the Pro and Ultra plans” or a model restrictionConnecting apps needs Pro or Ultra; upgrade, then reconnect. Otherwise check the account’s credits: a connected app can still request a model outside that account’s access.
Connections says app sign-in is not enabled yetThe sign-in service is switched off for now. Use an account key meanwhile; recreating keys will not change it.
429 or too many requestsWait a minute, then check the existing job; do not submit duplicates.
Job remains queuedUse its existing job ID. A queue status is not a failed render.
Completed job is missing from this libraryOpen your Library. Confirm which account you signed in with when connecting.

If a generation fails, keep its exact error and job ID when asking for help. Those identify the failed operation without sharing any sign-in or key. If a browser tab or assistant session closes after submission, return to the same job rather than assuming it was canceled. Before reconnecting, distinguish a failed tool request from a job that was accepted and is still running.

OpenModels MCP setup FAQ

Can I use OpenModels MCP on the Free plan?

Not at the moment: connecting apps is part of the Pro and Ultra plans. On those plans each generation spends your own credits and follows your plan’s model access, so the models an assistant can run for you are the ones your account can run on the site.

Do I need an API key or client ID?

No. Claude, ChatGPT, Claude Code and other apps with sign-in register themselves when you paste the URL; you only sign in and press Connect. Account keys are only for clients that cannot open a sign-in window.

Where do images and videos created through MCP appear?

Jobs and saved generation assets belong to the OpenModels account you signed in with when connecting. Keep the job ID to retrieve status, and use the returned asset link or your account library to open the result.

How do I disconnect an app?

Open Connections and choose Disconnect next to the app. It loses access to your account straight away; connect it again from the app whenever you want.

What happens if I lose or expose an MCP key?

Create a replacement in Connections, update the client and revoke the old entry. The full original secret cannot be recovered from the key list. Use separate labelled keys when you want to revoke one client without changing every connection.

Open Connections — Opens the studio. Nothing runs, and nothing is charged, until you press Generate.

Last checked 2026-09-26. Prices and plan limits change; the pricing page is the authority.