Skill Function MCP Setup
for OpenCode.

Configure InferX Model

Open your OpenCode configuration file:

nano ~/.config/opencode/opencode.json

Add the InferX provider block — replace the placeholders with your actual tenant ID, model endpoint, and API key:

{
  "provider": {
    "inferx": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "inferx",
      "options": {
        "baseURL": "https://model.inferx.net/funccall/<TENANT_ID>/endpoints/<MODEL_ENDPOINT>/v1",
        "apiKey": "<INFERX_API_KEY>"
      },
      "models": {
        "Qwen/Qwen3-Coder-Next-FP8": {
          "name": "Qwen/Qwen3-Coder-Next-FP8",
          "limit": {
            "context": 262000,
            "output": 16384
          }
        }
      }
    }
  }
}

Configure InferX MCP Server

In the same file, add the MCP block alongside your provider config. Replace <INFERX_MCP_TOKEN> with your token from the console:

{
  "mcp": {
    "inferx": {
      "type": "remote",
      "transport": "streamable-http",
      "url": "https://model.inferx.net/mcp",
      "enabled": true,
      "headers": {
        "Authorization": "Bearer <INFERX_MCP_TOKEN>"
      }
    }
  }
}

Save the file.

Restart OpenCode

Apply the configuration by restarting OpenCode:

pkill -f opencode
opencode

Verify Connection

Ask OpenCode to list available tools:

If InferX tools appear in the list, the MCP connection is active. If you only see built-in tools like bash, read, or write, the MCP connection is not active — restart OpenCode and check that the MCP block is saved correctly.

Test Tool Execution

Confirm tools are being invoked automatically:

For multi-skill orchestration:

The model should invoke multiple InferX tools, synthesize the results, and report which tools were used.