Skip to content

Connecting MCP Clients

Any MCP-compatible AI tool can connect to the Workbench's MCP server endpoint. External clients gain access to the same tools and resources that power the built-in Ask Ellie assistant. The MCP server uses HTTP transport with Bearer token authentication.

Meet the following requirements before configuring a client.

  • The Workbench server must be running and accessible from the machine where the MCP client operates.
  • Authentication requires an API token. Create tokens through the admin panel under Security > Tokens, or use the REST API. See Users & Authentication for details on token management.

Endpoint

The MCP server exposes a JSON-RPC 2.0 endpoint at /mcp/v1 on the server's HTTP address.

In the following example, the endpoint URL uses the default server address:

http://localhost:8080/mcp/v1

Include the token in the Authorization header using the Bearer scheme. The token's scope controls which connections and MCP tools the client can access.

In the following example, a curl command sends a request to the MCP endpoint:

curl -X POST http://localhost:8080/mcp/v1 \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Replace the URL and token with values that match your environment.

Client Configuration

The following sections describe how to configure popular MCP clients. Each example connects to http://localhost:8080/mcp/v1 with a Bearer token. Replace the URL and token value with your own.

Claude Code

Claude Code stores MCP server configuration in ~/.claude.json for user scope or .mcp.json for project scope.

In the following example, the claude mcp add command registers the Workbench server:

claude mcp add ai-dba-workbench \
  http://localhost:8080/mcp/v1 \
  -t http \
  -H "Authorization: Bearer YOUR_TOKEN"

Alternatively, create a .mcp.json file in the project root for project-scoped configuration.

In the following example, the .mcp.json file uses an environment variable for the token:

{
  "mcpServers": {
    "ai-dba-workbench": {
      "type": "http",
      "url": "http://localhost:8080/mcp/v1",
      "headers": {
        "Authorization": "Bearer ${AI_DBA_WORKBENCH_TOKEN}"
      }
    }
  }
}

Set the AI_DBA_WORKBENCH_TOKEN environment variable in your shell before launching Claude Code.

Cursor

Cursor stores MCP server configuration in ~/.cursor/mcp.json for user scope or .cursor/mcp.json for workspace scope.

In the following example, the configuration file uses the ${env:VAR} syntax for the token:

{
  "mcpServers": {
    "ai-dba-workbench": {
      "type": "http",
      "url": "http://localhost:8080/mcp/v1",
      "headers": {
        "Authorization": "Bearer ${env:AI_DBA_WORKBENCH_TOKEN}"
      }
    }
  }
}

Set the AI_DBA_WORKBENCH_TOKEN environment variable in your shell before launching Cursor.

VS Code (GitHub Copilot)

VS Code stores MCP server configuration in .vscode/mcp.json at the workspace level. The top-level key is servers rather than mcpServers.

In the following example, the configuration file uses the ${input:name} syntax for the token:

{
  "servers": {
    "ai-dba-workbench": {
      "type": "http",
      "url": "http://localhost:8080/mcp/v1",
      "headers": {
        "Authorization": "Bearer ${input:ai-dba-workbench-token}"
      }
    }
  }
}

VS Code prompts for the input variable value when the MCP client connects. You can also use environment variables as an alternative to interactive input.

Windsurf

Windsurf stores MCP server configuration in ~/.codeium/windsurf/mcp_config.json.

In the following example, the configuration file uses serverUrl instead of url:

{
  "mcpServers": {
    "ai-dba-workbench": {
      "serverUrl": "http://localhost:8080/mcp/v1",
      "headers": {
        "Authorization": "Bearer ${env:AI_DBA_WORKBENCH_TOKEN}"
      }
    }
  }
}

Set the AI_DBA_WORKBENCH_TOKEN environment variable in your shell before launching Windsurf. Note that Windsurf uses serverUrl instead of url in the configuration.

Claude Desktop (Not Supported)

Claude Desktop does not support HTTP transport for MCP servers. The claude_desktop_config.json file only accepts stdio transport for locally installed MCP servers. Use Claude Code instead for connecting to the Workbench.

Verification

After you configure the client, it should discover the Workbench's tools automatically. Verify the connection by asking your AI assistant to list the available MCP tools or to run a simple query such as listing database connections.

If the client does not discover the tools, confirm that the server is running, the URL is correct, and the token is valid.

Available Tools

The Using Workbench with MCP Tools page documents the full list of tools and resources.

  • Ask Ellie describes the built-in AI assistant that uses these tools internally.
  • Using AI Features covers AI-powered summaries of database health and status.