ChatGPT can now connect with MCP servers, making the data, APIs, and other tools you use every day available in your ChatGPT sessions. This guide will walk you through the steps to build and deploy an MCP server to use as a ChatGPT Connector.
- Choose a template to get started.
- Once the deployment is created, copy the URL. Your MCP server is ready for a client to connect at the
/mcproute. - Now, follow ChatGPT’s documentation to create a connector. Enable “Developer Mode” in your Advanced settings, and then add your MCP Server in the Connectors tab.
Your MCP server is now available in your ChatGPT session.
- Start a new chat on ChatGPT.
- Enable Developer Mode.
- Click on “Add sources” and select your MCP server.
- Ask ChatGPT "Use the roll_a_die tool to get a number" to run the tool in the MCP server.
- On ChatGPT’s response, the tool implemented by the MCP will be called.
MCP servers give AI models access to external capabilities. By connecting ChatGPT to an MCP server, your chat in ChatGPT quickly becomes an agent that is able to interact with the tools you use every day.
For example, you can connect ChatGPT to Vercel using Vercel’s own MCP server. Once authenticated, ChatGPT can interact with your projects, deployments, and fetch documentation to help you build software.
Optimized cost: Fluid compute intelligently re-uses existing resources. This means Vercel Functions can be billed at Active CPU pricing, so you only pay for the compute you use. MCP servers benefit from this model due to their long idle times and quick message bursts.
Deployment previews: Preview deployments allow you to test out your changes before going live. To try out your MCP server, you can simply change the URL in ChatGPT and try them out.
Instant Rollbacks: If you deploy to production and don’t like the results, you’re able to roll back to a previous version of your MCP server in milliseconds. This is great for recovering from bugs, breaking changes, or unintended changes to your product.
Secure by default: Vercel Firewall is a best-in-class, multi-layered protection layer with a platform-wide firewall to automatically protect from abnormal traffic, like DDoS attacks, and web application firewall to define custom rules for your specific needs.
Vercel is the unified platform for AI workloads. Whether you're shipping an MCP server, agents, running generated code, or analyzing bot traffic to your existing application, we're here to help.