Vetrix Docs

MCP

The MCP settings page connects AI assistants to a single repository over the Model Context Protocol. From it you turn the repository's MCP server on or off, copy the server URL, create an access token for the assistant to authenticate with, and copy a ready-made configuration snippet for the client you use. Local clients authenticate with a personal access token; remote connectors can sign in with OAuth instead. Turning the server on or off requires repository administrator access.

Where to find it

  • MCP: gitvetrix.com/{owner}/{repo}/settings/mcp
  • Navigation: Open a repository, select Settings, then select MCP in the left settings sidebar.

Copy the server URL

The Server URL section shows the address AI assistants connect to for this repository. Select Copy MCP server URL to copy it. When no URL is available yet, the field reads Not configured.

Enable or disable the MCP server

The MCP server section controls whether assistants can reach this repository.

  1. Select the Enable the MCP server for this repository checkbox to turn it on, or clear it to turn it off.

Only a repository administrator can change this setting. If you are not an administrator, the checkbox is read-only and a note reads that only a repository administrator can change it. If MCP has been turned off for the whole instance by an administrator, a note explains that enabling it for this repository has no effect until MCP is re-enabled instance-wide.

Create an access token

Local MCP clients authenticate with a personal access token that carries the mcp:read scope.

  1. In the Authentication section, select Create MCP token for this repo. This opens the personal access token page with the scope and a suggested name already filled in.
  2. Create the token and copy it. The token is shown only once.
  3. Paste it into the client configuration snippet where the placeholder token appears.

See Personal access tokens for how tokens are created, scoped, and revoked.

Configure a local client

The Client configuration section builds a configuration snippet for the assistant you use.

  1. Under MCP client, choose your client: Claude Desktop, Cursor, VS Code, or Continue.dev.
  2. The page shows the configuration file to add the snippet to and generates the matching JSON, already pointing at this repository's server URL.
  3. Copy the snippet into that file, replacing the token placeholder with the token you created.

Each snippet includes an Authorization header that carries your token, so the assistant connects as you.

Connect a remote AI assistant with OAuth

The Remote (OAuth) section is for remote connectors that sign in with OAuth instead of a personal access token. What it shows depends on whether the instance offers automatic client registration.

  • When automatic registration is available, a zero-config path appears: add the server URL shown in this section to your connector, and it registers itself and prompts you to approve the mcp:read scope. No client ID is needed.
  • When it is not, use a pre-registered OAuth client. Register a dedicated OAuth application first, then paste its client ID (and secret, where the connector requires one) into the connector.

To set up a pre-registered client:

  1. Under Step 1 — Create an OAuth application to get a Client ID, select Create an OAuth application and register one. Set its redirect URIs to the values shown on the page and request the mcp:read scope. The scope and any required redirect URIs are shown with copy buttons.
  2. Under Remote OAuth connector, choose your connector: Claude custom connector, Gemini CLI, or OpenAI.
  3. The page shows where to paste the generated snippet and lists any redirect URIs an administrator must register on the OAuth client. A connector that uses a bring-your-own access token needs no redirect URI — mint an mcp:read token and paste it into the connector's Authorization header.

Do not reuse a "Sign in with Google" or other single sign-on client ID here. It is not a Vetrix OAuth application, and a connector rejects it. Create a dedicated OAuth application in Step 1 and use that client ID.

For a full walkthrough, follow the link to the guide on connecting an AI assistant over OAuth in this section of the page.

Tips

  • If the MCP settings fail to load, the page shows Couldn't load MCP server settings with a retry control in place of the server URL and toggle. Retry once the connection is restored.
  • Automatic client registration is an instance-wide setting an administrator controls. If your connector needs it but the instance does not offer it, ask an administrator to enable it, or use the pre-registered client path in the meantime.