Agent Mode - ModRouter

Overview

Agent Mode lets ModRouter run selected MCP tools automatically when a model requests them. This is useful for workflows where the model should call tools, receive results, and continue until it can produce a final answer. When Agent Mode runs:

The model returns one or more tool calls.
ModRouter executes tools that are configured for auto-execution.
Tool results are sent back to the model.
The loop continues until the model returns a final answer or the maximum depth is reached.

Agent Mode is not compatible with streaming responses. Use non-streaming chat or responses endpoints when auto-executing tools.

Configure Auto-Executable Tools

Use tools_to_execute to make tools available and tools_to_auto_execute to allow selected tools to run without application-side approval.

{
  "mcp": {
    "client_configs": [
      {
        "name": "filesystem",
        "connection_type": "stdio",
        "stdio_config": {
          "command": "npx",
          "args": ["-y", "@anthropic/mcp-filesystem"]
        },
        "tools_to_execute": ["*"],
        "tools_to_auto_execute": ["read_file", "list_directory"]
      }
    ],
    "tool_manager_config": {
      "max_agent_depth": 10,
      "tool_execution_timeout": "30s"
    }
  }
}

Only tools included in tools_to_execute can be auto-executed.

Send a Request

export MODROUTER_BASE_URL="https://api.modrouter.online"
export MODROUTER_API_KEY="<your-modrouter-virtual-key>"

curl --location "https://api.modrouter.online/v1/chat/completions" \
  --header "Authorization: Bearer ${MODROUTER_API_KEY}" \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "openai/gpt-4o-mini",
    "messages": [
      { "role": "user", "content": "Read the project README and summarize it." }
    ],
    "tool_choice": "auto"
  }'

Go SDK

mcpConfig := &schemas.MCPConfig{
    ClientConfigs: []schemas.MCPClientConfig{
        {
            Name:           "filesystem",
            ConnectionType: schemas.MCPConnectionTypeSTDIO,
            StdioConfig: &schemas.MCPStdioConfig{
                Command: "npx",
                Args:    []string{"-y", "@anthropic/mcp-filesystem"},
            },
            ToolsToExecute:     []string{"*"},
            ToolsToAutoExecute: []string{"read_file", "list_directory"},
        },
    },
    ToolManagerConfig: &schemas.MCPToolManagerConfig{
        MaxAgentDepth: 10,
    },
}

Choosing Auto-Executable Tools

Only mark deterministic, low-risk tools for auto-execution. For tools that write files, trigger external actions, or make irreversible changes, return the tool call to your application and ask for approval before execution.