3 Agentic AI Issues

Review Oracle Integration Agentic AI issues.

Topics:

MCP Server URL for a Project is Incorrect

Issue: When you enable MCP Server for a project, the displayed URL is incorrect.

Workaround: Construct your MCP server URL in the following format: 

https://<NameOfServiceInstance>.integration.<region>.ocp.oraclecloud.com/mcp-server/v1/projects/<ProjectName>/mcp

Example:

https://myinstance.integration.eu-frankfurt-1.ocp.oraclecloud.com/mcp-server/v1/projects/PROJECTA/mcp

AI Agent Activation Fails with Project Deployment

Issue: If you use a project deployment to export AI agents and the related integrations to another Oracle integration instance, AI agent activation fails.

Workaround: On the new Oracle integration instance, create a new AI agent and activate it.

Prompt Templates Created Prior to 26.04 Have Extra Characters on Import

Issue: When you import a project or deployment with a prompt template created prior to version 26.04, extra characters are added to the prompt template. For example: "\"\\\"Add two numbers {{data.num1} and {{data.num2}}\\\"\"".

Workaround: If you created prompt templates prior to Oracle Integration 26.04, delete those prompt templates and recreate them in version 26.04.

Conversation ID Does not Work with AI Agents Created Prior to 26.04

Issue: When an AI agent is invoked with an API or through the AI agent endpoint, conversation ID does not work. You cannot start a conversation with a specific conversation ID or continue a conversation with a specific conversation ID.

Workaround: Specify the conversation ID with the query parameter conversationId in the AI agent endpoint. For example: 

https://<host>/ic/api/integration/v2/flows/rest/aiagent/project/<projectCode>/<agentCode>/<version>/start?conversationId=444444445555555
Body payload: {"prompt":"what is the status of order 100?","conversationId":"444444445555555"}

Invoking an AI Agent with Incorrect URL Creates Conversation and AI Agent Instance Record

Issue: If you specify an incorrect URL to invoke an AI agent, the agent is not started and a 404 is the response. However, a conversation record and AI agent instance record will be created. As a result, the AI agent instance status remains In Progress and the integration status is Errored.

Workaround: When invoking an AI agent, ensure the path you specify is correct and validate the URL before invoking the AI agent.

Calling Integrations that Include the RAG search or RAG ingest Actions Using the REST API

Calling integrations that include the RAG search and RAG ingest actions with the REST API requires the caller to use one of the following forms of security access: