Connect an AI client
Give an AI client access to published queries or authorized modeling tasks through Model Context Protocol (MCP).
Enable the service you need
In Runtime Settings → MCP services (运行配置 → MCP 服务), an administrator can enable query and modeling independently. Both default to disabled. Enable publication separately if it is required.
/mcp/query- Discover visible models and published schema, then run bounded read-only queries.
/mcp/modeling- Read, validate, create, and save authorized drafts. Publication requires the publisher role and server opt-in.
Register GraphLake in your client
Open AI in the console. Generate a personal token and store it in the client's credential configuration. Add a Streamable HTTP MCP entry for each service you need. Replace the example origin with your GraphLake HTTPS origin and provide the token in the Authorization header.
{
"mcpServers": {
"graphlake-query": {
"url": "https://graphlake.example/mcp/query",
"headers": {
"Authorization": "Bearer <PERSONAL_AI_TOKEN>"
}
}
}
}The token uses the owner's current model permissions and tenant scope. An administrator can issue a token for a member with MCP access. Test the connection in the AI workspace to check the authenticated identity and tool visibility.
Run your first request
Ask the client to list permitted models, inspect the published schema, and run a small query. For model edits, read the current draft and revision, validate changes, and review the saved draft before publishing. See the MCP reference for role checks, token rotation, and query limits.