Ollama remote api not working and provider diagnostic shows incorrect API base URL when custom api_base is configured

[ricardoamaro]

Bug Description

The mcp-cli provider diagnostic ollama command shows the default http://localhost:11434 API base URL even when a custom api_base is configured in ~/.chuk_llm/providers.yaml.

Steps to Reproduce

Verified the configuration is saved:

cat ~/.chuk_llm/providers.yaml

Shows:

ollama:
  api_base: http://192.168.0.22:11434
  default_model: mistral

3. Run diagnostic:

uvx mcp-cli provider diagnostic ollama

Expected Behavior

The diagnostic should display:

│ ollama   │ ✅ Ready │ 7 models │ 📡       │ API: http://192.168.0.22:11434 | Discovery: ✅ │

Actual Behavior

The diagnostic shows:

│ ollama   │ ✅ Ready │ 7 models │ 📡       │ API: http://localhost:11434 | Discovery: ✅ │

Prompt Diagnostics

$ curl http://192.168.0.22:11434
Ollama is running  

$ uvx mcp-cli cmd --provider ollama --model mistral --prompt "hello"
✓ MCP CLI ready
Available commands: chat, cmd, models, ping, prompts, provider, providers, resources, servers, tools
Use --help to see all options
Secure MCP Filesystem Server running on stdio
Allowed directories: [ '/projects/hostdir' ]
[07/13/25 02:19:00] INFO     Processing request of type PingRequest                                                                                            server.py:558
INFO     Processing request of type ListToolsRequest                                                                                       server.py:558
[07/13/25 02:19:00] INFO     Processing request of type PingRequest                                                                                            server.py:558
INFO     Processing request of type ListToolsRequest                                                                                       server.py:558
ERROR    Error in Ollama completion: Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download
Error: Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download

$ uv pip install mcp-cli --system
Using Python 3.13.4 environment 
Resolved 53 packages in 20ms
Installed 2 packages in 15ms
 + chuk-mcp==0.5
 + mcp-cli==0.2.4

Possible Root Cause

The provider info appears to contain the library's default values rather than the user's customized configuration from ~/.chuk_llm/providers.yaml, as per the documentation.

[ricardoamaro]

Looking at the code in src/chuk_llm/configuration/unified_config.py, the library specifically looks for these files in this order:

1. File specified by the CHUK_LLM_CONFIG environment variable
2. chuk_llm.yaml in the current working directory (REPLACES package config)
3. providers.yaml in the current working directory (EXTENDS package config)
4. llm_config.yaml in the current working directory (EXTENDS package config)
5. config/chuk_llm.yaml in the current working directory
6. ~/.chuk_llm/config.yaml in the home directory

So I just did:

$ mv ~/.chuk_llm/providers.yaml ~/.chuk_llm/config.yaml

And now it seems to work:

$ uvx mcp-cli cmd --provider ollama --model mistral --prompt "just say hello"
✓ MCP CLI ready
  Available commands: chat, cmd, models, ping, prompts, provider, providers, resources, servers, tools
  Use --help to see all options
WARNING  Config file 'server_config.json' not found.

 Hello! I'm here to help you with your questions or tasks. Let me explain how we can work together effectively.....

Will file PR to update the docs.

[chrishayuk]

thanks for doing this.. appreciate it