update for ollama backend

commons/mcp_manager.py

@@ -1,3 +1,4 @@
+import os
 import time
 import threading
 import asyncio
@@ -92,13 +93,15 @@ class MCPHostManager:
 
     def _build_command(self) -> List[str]:
         """Build the command to start mcphost"""
+        os.environ["OLLAMA_HOST"] = settings.ollama_host
+        os.environ["OLLAMA_NUM_CTX"] = str(settings.ollama_num_ctx)
         command = [
             settings.mcphost_path,
             '--config', settings.mcphost_config,
             '--model', settings.mcphost_model,
-            '--openai-url', settings.openai_url,
-            '--openai-api-key', settings.openai_api_key
         ]
+        # '--openai-url', settings.openai_url,
+        # '--openai-api-key', settings.openai_api_key
 
         if settings.debug:
             command.insert(1, '--debug')

test.sh

@@ -8,7 +8,7 @@
 #  http://localhost:8000
 
 curl -X 'POST' \
-  'http://0.0.0.0:8000/v1/chat/completions' \
+  'http://10.8.0.10:33759/v1/chat/completions' \
   -H 'accept: application/json' \
   -H 'Content-Type: application/json' \
   -d '{
@@ -16,10 +16,10 @@ curl -X 'POST' \
   "messages": [
     {
       "role": "user",
-      "content": "can you give me your previous answer in JSON format? /no_think"
+      "content": "sum all first 10 digits of pi? /no_think"
     }
   ],
   "temperature": 0.7,
   "stream": false,
-  "max_tokens": 1024
+  "max_tokens": 2058
 }'