feat: MCP/OpenAPI tool calling support

steward/main.py

@@ -10,6 +10,7 @@ from steward.config import get_settings
 from steward.llm.client import LLMClient
 from steward.memory.thread_store import ThreadMemoryStore
 from steward.proposals.generator import ProposalGenerator
+from steward.tools.client import ToolClient
 
 logging.basicConfig(
     level=logging.INFO,
@@ -46,7 +47,18 @@ def main() -> None:
 
     llm = LLMClient(settings)
     thread_store = ThreadMemoryStore(settings.thread_memory_path)
-    app = build_application(settings, llm, thread_store)
+
+    tool_client: ToolClient | None = None
+    if settings.mcp_server_url:
+        tool_client = ToolClient(
+            base_url=settings.mcp_server_url,
+            api_key=settings.mcp_server_api_key,
+        )
+        logger.info("MCP tool server configured: %s", settings.mcp_server_url)
+    else:
+        logger.info("No MCP_SERVER_URL configured – tool calling disabled")
+
+    app = build_application(settings, llm, thread_store, tool_client)
     generator = ProposalGenerator(settings, llm)
 
     scheduler = AsyncIOScheduler()
@@ -60,9 +72,10 @@ def main() -> None:
     scheduler.start()
 
     logger.info(
-        "Steward starting: model=%s analysis_url=%s",
+        "Steward starting: model=%s analysis_url=%s tools=%s",
         settings.openai_model,
         settings.analysis_target_url or "(none)",
+        settings.mcp_server_url or "(none)",
     )
     app.run_polling(allowed_updates=["message"])