Connecting to MCP manually

deepmirror exposes a remote Model Context Protocol (MCP) server, so you can train models, run predictions, and design molecules from Claude, OpenAI Codex, or your own application.

You don't need separate MCP servers. Register the same remote endpoint in each client:

https://api.app.deepmirror.ai/mcp

Claude

For Claude.ai or Claude Desktop:

  1. Open Settings → Connectors.
  2. Select Add custom connector.
  3. Name it deepmirror.
  4. Paste the URL above.
  5. Select Add, then Connect if authentication is requested.

Custom remote connectors require Claude Pro, Max, Team, or Enterprise. Team and Enterprise connectors may need to be added by an organisation owner. See Anthropic's connector guide.

To upload files (for example a CSV or SDF to create a dataset), Claude must be allowed to send network requests to api.app.deepmirror.ai. If your Claude environment restricts outbound network access, add api.app.deepmirror.ai to its allowlist, otherwise file uploads will fail even though the connector is connected.

For Claude Code:

claude mcp add --transport http --scope user deepmirror https://api.app.deepmirror.ai/mcp
claude mcp list

Inside Claude Code, run /mcp to authenticate or inspect the connection. See the Claude Code MCP documentation.

OpenAI Codex

In the Codex desktop app:

  1. Open Settings → Plugins.
  2. Select MCP, then Add.
  3. Name it deepmirror.
  4. Choose Streamable HTTP.
  5. Enter the URL.
  6. Save and restart Codex.
  7. Select Authenticate if prompted.

Or use the terminal:

codex mcp add deepmirror --url https://api.app.deepmirror.ai/mcp
codex mcp list

If authentication is required:

codex mcp login deepmirror

Codex desktop, CLI, and its IDE extension share this configuration. See the OpenAI MCP documentation.

OpenAI Responses API

If you are building your own application on the OpenAI API, add this tool to a Responses API request:

{
  "type": "mcp",
  "server_label": "deepmirror",
  "server_url": "https://api.app.deepmirror.ai/mcp",
  "require_approval": "always"
}

Keep approval enabled initially, especially for tools that create datasets, train models, or otherwise modify data. See the OpenAI MCP API guide.

Authentication

The URL alone is sufficient when the client supports a standard OAuth flow. If a client reports 401 Unauthorized, you'll also need an OAuth client configuration or bearer token from deepmirror.

Privacy

See the deepmirror MCP privacy policy for how data sent through the MCP server is handled.