Federico Mariani created CAMEL-23397:
----------------------------------------
Summary: camel-openai: Support tool annotations/hints for
returnDirect and safety classification
Key: CAMEL-23397
URL: https://issues.apache.org/jira/browse/CAMEL-23397
Project: Camel
Issue Type: Improvement
Components: camel-openai
Reporter: Federico Mariani
Extends [CAMEL-23382|https://issues.apache.org/jira/browse/CAMEL-23382]
The MCP integration in _camel-openai_ already supports rich tool metadata via
MCP tool annotations:
- _returnDirect_: short-circuit the agentic loop and return the tool result
directly to the caller
- `readOnlyHint`, `destructiveHint`, `openWorldHint`: safety classification
hints
However, these annotations are only available for MCP-sourced tools. When
[CAMEL-23382|https://issues.apache.org/jira/browse/CAMEL-23382] introduces
route-based tools via `AiToolSpec`, those tools will lack this metadata, making
them less capable than MCP tools used through the same agentic loop.
*Proposed changes:*
The `AiToolSpec` defined in CAMEL-23382 should include an extensible
annotations mechanism:
- `returnDirect` (boolean, default false) — when true, the agentic loop returns
the tool result directly without sending it back to the model. This is critical
for tools that produce final output (e.g., "generate report", "fetch file").
The agentic loop should treat annotations uniformly regardless of whether the
tool came from MCP or from the Camel route-based registry. This means the
`AiToolRegistry` lookup should return annotations alongside the tool spec.
*Example usage:*
```yaml
- from:
uri: "ai-tools:generateReport"
parameters:
description: "Generate a PDF report from the given data"
returnDirect: true
steps:
- to: "pdf:create"
```
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
