davsclaus opened a new issue, #3362: URL: https://github.com/apache/camel-kamelets/issues/3362
A new source Kamelet, `ai-tool-source`, on the Camel `ai-tool` component (Preview since 4.22): it offers the flow after it as a tool that an AI agent can call. With `camel-mcp-server`, any MCP client sees the tool, such as Claude Desktop or Cursor. The `openai` component can call it too. - **Options:** `toolName`, `description`, `argSchema` (the JSON Schema of the arguments, inline or from `classpath:`/`file:`), and `tags` (required). The `readOnlyHint`, `destructiveHint`, `idempotentHint` and `openWorldHint` hints are optional, and so is `allowedRoles`. - **Output:** the body is the arguments of the call as JSON text (`argumentsAsBody`), without undeclared arguments. The reply of the flow is the result of the tool. - **Security:** an MCP client is an untrusted sender. MCP only exposes tools whose tag is in `camel.server.mcp.tags`, so `tags` is required and nothing is exposed by accident. The description explains this setting and `allowedRoles`. - **Citrus tests:** an integration with `camel-mcp-server`. The test calls `tools/list` (the tool is there with its schema) and `tools/call` (the flow gets the arguments and its reply is the result) over HTTP. It also checks that a tool with another tag is not listed. _Claude Code on behalf of Claus Ibsen_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
