rpa-platform-agent-mcp

rpa-platform-agent-mcp is the stdio launcher used by the RPA training platform to expose task tools through MCP.

Students and teachers only need to add one MCP config. The launcher handles first-use authorization automatically and then reuses the cached login state.

Production usage

Add this config to any AI tool that supports stdio MCP:

{
  "mcpServers": {
    "rpa-platform-agent": {
      "command": "npx",
      "args": [
        "-y",
        "rpa-platform-agent-mcp",
        "--base-url",
        "https://your-domain.example/mcp"
      ]
    }
  }
}

First use

On first use, the launcher may open the browser once to complete authorization.

After authorization succeeds, the MCP access token is cached locally and reused automatically.

What users should do next

1. Save the MCP config in the AI tool.
2. Reopen the AI tool.
3. Confirm that rpa-platform-agent appears in the MCP server list.
4. Start with:

get_task_context({
  "scenarioCode": "S05",
  "taskId": "task_xxx"
})

Local cache

By default, the launcher stores the cached access token at:

~/.rpa-platform-agent/mcp-access-token.json

Optional arguments

• --base-url <url>: required MCP endpoint base URL.
• --cache-file <path>: override the local token cache file.
• --token-file <path>: optional fallback token file for local development only.

Development

For repository-local development, the platform can still fall back to the local package entry instead of npx. That path is intended for developers, not for student-facing docs.