Using Gemini CLI with SoAI

Gemini CLI can connect to SoAI's Streamable HTTP MCP server. This gives the client access to the tools that an administrator exposes. The custom base URL in Gemini CLI speaks the Gemini API rather than the OpenAI-compatible protocol, so this connection is MCP-only.

Configure and verify MCP access

  1. In the SoAI WebUI, select the Settings button at the right of the header, choose Settings, and open MCP > MCP tokens.
  2. The MCP tab appears only to an Administrator with MCP administration (MCP_ADMIN). Create a personal MCP token and copy it when shown; token creation also requires MCP usage (MCP_USE).
  3. Enabling the MCP runtime and server mode and selecting the tools, resources, and prompts exposed to remote clients also requires Configuration changes (CONFIG_PATCH).

Add the server

Merge this entry into the user or trusted-project Gemini settings:

{
  "mcpServers": {
    "soai": {
      "url": "https://SOAI_HOST:5090/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer ${SOAI_MCP_TOKEN}"
      }
    }
  }
}
export SOAI_MCP_TOKEN='YOUR_MCP_TOKEN'
export NODE_EXTRA_CA_CERTS='/path/to/soai-ca.pem'
gemini mcp list
gemini

Open Gemini CLI inside a workspace you trust through its normal folder-trust prompt. Enter /mcp to open the interactive MCP status panel, confirm that soai and its tools are connected, then ask it to use a harmless exposed read-only tool. A saved list entry alone is not sufficient evidence.

Security and troubleshooting

Do not disable folder trust globally and do not disable TLS verification. If the server is configured but no tools appear, check the separate server-mode switch, exposed-tool list, MCP token, and /mcp URL. Gemini's model authentication remains independent from the MCP token.

Related documentation