Summary

• Centralize tool management: Move MCP tool loading and RAG tool preparation from AIChatCompletionIntegrationService and AIManagementChatClient<T> into AIManagementChatClient.PrepareOptionsAsync(), creating a single responsibility point for tool management during chat completions.
• Add HasEmbedderConfiguration flag to ChatClientCreationConfiguration so RAG tools are only added when an embedder provider is actually configured for the workspace, preventing unnecessary tool injection.
• Simplify AIManagementChatClient<T> to a thin resolver wrapper that delegates directly to the resolved IChatClient, removing duplicated MCP tool loading and FunctionInvokingChatClient wrapping logic.
• Fix MCP argument parsing in both McpServerAppService and McpServerConfigurationStore to support space-separated argument strings (e.g. -y @modelcontextprotocol/server-filesystem /tmp) in addition to JSON arrays.
• Increase default MCP timeout from 30s to 120s and properly pass it as InitializationTimeout to both HttpMcpClientFactory and StdioMcpClientFactory.
• Fix duplicate [Route] attribute on ChatCompletionsController (removed extraneous [Route("v1")]).
• Add unit tests for AIManagementChatClient covering RAG+MCP tool merging, MCP-only tools, FunctionInvokingChatClient validation, and caller-provided tool preservation.

Test plan

• Add AIManagementChatClient_Tests with 4 test cases:
  • RAG + MCP tools are both added when embedder is configured
  • Only MCP tools are added when embedder is not configured
  • Exception is thrown when RAG is enabled but inner client is not FunctionInvokingChatClient
  • Caller-provided tools are preserved when merging with RAG and MCP tools

10.2.0-rc.1 24 February, 2026 abp-module-ai-management