Add a Custom MCP Connector

Updated November 2, 2026 — Le Chat is now Vibe. As of May 28, 2026, Le Chat has been rebranded as Vibe (still at chat.mistral.ai): a single agent for work and code, carrying over your conversations, settings, and subscriptions. The legacy Le Chat experience remains available in Chat mode; in Work mode, agents are now skills (invoked with @), memories serve as the knowledge base, and deep search is the Deep Research skill. Plans are now called Mistral Free, Pro, Team, and Enterprise. This lesson uses the name Vibe; some menus described here may have changed names or locations.

Beyond the Official Directory

Built-in connectors cover many use cases, but your organization likely relies on internal tools, proprietary APIs, or niche services absent from the Vibe directory. This is where custom connectors come in: you can add any remote MCP server to Vibe simply by providing its URL. That internal CRM built six years ago becomes usable in a conversation, just like GitHub or Notion.

The Addition Process

Adding a connector takes just a few clicks in the interface. Open the Connectors tab, click Add a custom connector at the bottom of the list, and fill in the required details:

• Name: A descriptive name to identify the connector (for example, “Internal CRM” or “Weather API”)
• Server URL: The public address of your MCP server, including the /mcp endpoint (for example, https://mon-serveur.example.com/mcp)
• Description: A brief explanation of what the connector does (optional but recommended)
• Authentication method: None, Token, or OAuth 2.0

Finally, click Create to save the connector. Vibe automatically queries the server to discover available tools: the list of exposed functions appears, and you can start using them in your conversations.

This discovery request returns the name, text description, expected parameters (with their name, type, and description), and return type for each tool. These are the only details the Mistral model uses to decide when and how to invoke each tool—which is why the next lesson focuses heavily on description quality. A well-written tool with a poor description remains invisible to the model.

Authentication Methods

The simplest case is no authentication: your MCP server is accessible to anyone who knows its URL. This works for public tools that don’t handle sensitive data, prototypes, demos, and MCP servers wrapping public APIs. Token authentication involves providing a key sent with every request to the server; Vibe stores the token securely. Choose this for internal APIs protected by an API key, servers that need to identify the caller, and any scenario where you want to restrict access. Finally, there’s OAuth 2.0, the full flow with redirection to the service’s authorization page: it’s the most secure mechanism but also the most complex to implement on the server side, and it will be covered in depth in future Mistral AI guides.

Prerequisites for Your MCP Server

The public URL requirement might seem to block local development, but there’s a workaround during the debugging phase. If you’re developing on your machine and want to test with Vibe before deploying, open a tunnel to your local server using ngrok:

ngrok http 8000

ngrok gives you a temporary public URL that you can use as the connector URL in Vibe. This approach is ideal for development and testing: you modify your code, restart, and Vibe immediately sees the new version without any deployment.

Testing with the MCP Inspector

Before adding your server to Vibe, test it with the MCP Inspector, an open-source tool that verifies your server responds correctly to discovery requests, lists available tools with their descriptions and parameters, executes tool calls manually to check results, and helps debug formatting or typing issues. Launch it with a simple npx command:

npx @modelcontextprotocol/inspector

The Inspector opens a local web interface where you connect your MCP server and test each tool individually. This step isn’t just a formality: it saves you from burning conversation messages on debugging and clearly isolates server errors from model interpretation errors.

Organizational Management

If you’re part of an organization on Vibe, note that custom connectors you add are visible to other members. An administrator can disable any connector for the entire organization if they deem it inappropriate or insecure. However, each user must provide their own credentials—token or OAuth—for connectors requiring authentication: sharing a connector never means sharing access credentials.

Security of Custom Connectors

Unlike official connectors validated by Mistral AI, a custom connector can execute arbitrary code on the server side, and the text description you see guarantees nothing about what the server actually does. Stick to servers you’ve developed yourself, those from trusted colleagues or partners, and open-source servers whose code you’ve verified. Even with your own connectors, keep human validation enabled throughout the debugging phase; once you’re certain everything works, you can consider setting read-only tools to “Always Allow.”

Key Takeaways

• You can add any remote MCP server as a custom connector in Vibe
• The server must be accessible via a public URL with an /mcp endpoint
• Three authentication methods are supported: none, token, and OAuth 2.0
• Use ngrok to test your local server with Vibe before deploying
• The MCP Inspector is essential for debugging your server before integration
• Only add connectors from trusted sources and keep human validation enabled