Agent Interoperability best practices
- Updated: 2025/12/16
To make the most of Agent Interoperability, check out the best practices.
Important: This feature is not generally available and
is restricted to specific customers. Contact your Automation
Anywhere account team for more information.
- Design the Agent system prompt/persona to control how tools are used:
- If you want your Agent to always ask for user confirmation before using any tool, include this detail in your Agent system prompt/persona when designing the Agent.
- If you want your Agent to always call the GetAutomationResult tool after using the Run Automation tool or other automation tools, include this detail in your Agent system prompt/persona when designing the agent.
- Update the tool descriptions when creating Agent connections to improve how
those tools work:
- If you want your Agent to do something before or after using a tool, you can update the tool descriptions those tools.
- If you want your Agent to check the tool inputs for a specific format or limit, you can update the tool descriptions for those tools.
- Before you give any business task to your Agent, ask them to list the automation. For example, enter: Get me the list of automations.
- After setting up your MCP client, you have configured your Automation Anywhere MCP server. Make sure you can access all three default tools provided by Automation Anywhere: DiscoverAutomation, RunAutomation, and GetAutomationResult. If you can see these tools, it means your Automation Anywhere MCP server is configured successfully, and you can connect to it through this MCP client.
- Automation discovery, invocation, and results extraction rely heavily on the AI Model/LLM.
The way the LLM creates prompts is crucial for discovering, invoking automation
tools, and getting results. Based on our experience, Cloud models perform better
than GPT models:
- Recommended models: Claude Sonnet 4.5 and Gemini 2.5 Pro
- Models will low efficacy/reliability: GPT 4.1 or earlier