MCP servers · Replicate · connector · GPT action
Connect ChatGPT to Replicate
One line gives ChatGPT your Replicate account as tools: 8 endpoints, predictions, models, more, called with your own connection. No key in the prompt, a receipt for every call, revoke ChatGPT's access alone any time.
Three steps
- Create a free 0n3 account and connect Replicate. Paste a key or sign in; it lands in your vault, encrypted.
- Install 0n3 into ChatGPT. A custom connector in Developer mode, or an Action on a custom GPT from the OpenAPI spec.
- Ask ChatGPT: whoami. Then: "Create a prediction in Replicate".
Custom connector (needs Developer mode)
- Settings → Connectors (or Apps) → Advanced → Developer mode on
- Create → name 0n3 · MCP server URL https://0n3.app/mcp · Authentication: OAuth
- Finish the sign-in with your 0n account
- Enable the connector in a chat and ask: "whoami"
ChatGPT connectors cannot send a static header, so 0n3 signs you in through its OAuth door (dynamic client registration is on).
The line above carries a placeholder; 0n3.app/install mints a real key named for ChatGPT and fills it in.
What ChatGPT can then do in Replicate
Predictions · Run model predictions4
Create a prediction
POST /v1/predictionsGet a prediction
GET /v1/predictions/{prediction_id}List predictions
GET /v1/predictionsCancel a prediction
POST /v1/predictions/{prediction_id}/cancelModels · Browse and search models3
Get a model
GET /v1/models/{owner}/{name}List model versions
GET /v1/models/{owner}/{name}/versionsSearch models
GET /v1/modelsMore · Everything else this app exposes.1
List collections
GET /v1/collectionsReplicate in other AI clients
Claude Code + ReplicateMCP server + skillClaude (desktop & claude.ai) + ReplicateconnectorCodex CLI + ReplicateMCP serverGemini CLI + ReplicateMCP serverCursor + ReplicateMCP serverVS Code (Copilot) + ReplicateMCP serverWindsurf + ReplicateMCP server