Federico Mariani created CAMEL-24334:
----------------------------------------
Summary: camel-ai-tool - raw JSON Schema tool input (argSchema)
for nested parameters
Key: CAMEL-24334
URL: https://issues.apache.org/jira/browse/CAMEL-24334
Project: Camel
Issue Type: Improvement
Components: camel-ai
Reporter: Federico Mariani
Follow-up from CAMEL-24308 (camel-mcp-server).
The ai-tool consumer endpoint currently supports only the flat
{{parameter.NAME=type}} syntax (with {{.description}}/{{.required}}/{{.enum}}
refinements). Nested or complex input schemas — arrays, nested objects, oneOf —
are a common need for MCP tools and cannot be expressed today.
Proposed scope:
* New optional {{argSchema}} endpoint option accepting a raw JSON Schema object
(inline string and/or {{resource:}} reference), mutually exclusive with the
flat {{parameter.*}} syntax.
* {{AiToolParameterHelper}} keeps generating the schema for the flat syntax;
with {{argSchema}} the provided schema is used as-is for
{{AiToolSpec.getParametersJsonSchema()}}.
* *Security-critical*: {{AiToolExecutor}}'s argument allowlist currently
derives accepted argument names from {{parameterDefs}}. With a raw schema, the
allowlist MUST be derived from the schema's top-level {{properties}} —
otherwise every argument is dropped (or worse, the filter is bypassed). This is
the same bug class as CAMEL-24241; the required-parameter validation needs the
same treatment.
* Engines consuming the structured {{parameters()}} map (quarkus ToolManager
{{addArgument}}) need the schema-based path: quarkiverse
{{ToolDefinition.setInputSchema(Object)}} covers it; the MCP Java SDK and
Spring AI engines already consume the JSON Schema string directly.
* The langchain4j-agent adapter builds native schema objects from
{{ParameterDef}} — it needs a raw-schema mapping too ({{JsonObjectSchema}} from
JSON).
----
_This issue was drafted by Claude Code on behalf of Federico Mariani._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
