Poznote with Ollama AI?

[Stephane-30]

Is it possible to connect Ollama with Poznote MCP-Server? I tried but it's not working. I obviously changed the port.
There are no input prompts in PN like in the animation MCP-Server

[timothepoznanski]

Hi!

A few clarifications:

Poznote itself doesn't have a chat/prompt input, the input you see in the animation is the AI assistant's chat (VS Code Copilot in that demo), not Poznote's UI. Poznote's MCP server is only a bridge: it exposes your notes as tools at http://127.0.0.1:8045/mcp for an AI assistant to call.

Ollama alone can't connect to it, because Ollama is a model runtime, not an MCP client. To use your notes with a local Ollama model, you need an MCP-capable chat app that uses Ollama as its backend and supports HTTP (streamable-http) MCP servers for example Goose, mcphost, or oterm. Point that app at Ollama for the model and add http://127.0.0.1:8045/mcp as an MCP server.

Also make sure the Ollama model you pick supports tool calling (e.g. llama3.1, qwen3, mistral) otherwise the assistant will chat but never actually use the Poznote tools.

Setup docs for the supported clients are here: MCP Server documentation.

In the meantime, I'll be working on an integrated chat that will be able to plug into Ollama.

[Stephane-30]

Hi Tim, thanks for the clarification. I understand AI as a concept but I'm quite new to implementing it into my workflow. Anyway, I managed to hook PN to goose and it kind of works. Now I need to understand how to communicate with PN using Chat in goose.

[timothepoznanski]

Working on it. 😊

[timothepoznanski]

New version available. 😊

[Stephane-30]

Hi Tim, unfortunately I can't get it to work. It fails with:

Connection failed: Failed to connect to 127.0.0.1 port 11434 after 0 ms: Could not connect to server

I have installed Ollama locally on my Linux machine. If I run it and enter http://localhost:11434 or http://127.0.0.1:11434 in my browser it comes up with 'Ollama is running'. That means Ollama is working correctly. The same URL works fine with another app I have installed.

Anything I can do to locate the error?

[timothepoznanski]

Hi! Yes, that's expected. Your Ollama instance is listening only on localhost (which is the default configuration), but Poznote is running inside a Docker container. Since the container has its own network namespace, it cannot reach Ollama through localhost. In other words, they're on two different local networks and can't communicate by default.

This is explained in the documentation, but I realized it wasn't as clear as it could be, so I've updated it with a more detailed explanation.

Please see: https://github.com/timothepoznanski/poznote#ai-assistant, then click on AI Assistant documentation.

[Stephane-30]

Okay Tim @timothepoznanski , that has explained a lot and I understand it much better now. It didn't help either that my Ollama was installed on my workstation (not docker) and my PN runs as a docker container on my little server. My workstation is quite powerful but my server is just a small mini PC with only 4 processors.
I've installed Ollama as a docker container with open webui on my server and it connected without any issues to my PN instance. The only problem is that my server starts sweating when I use a simple AI instruction.
For the moment I will have to work without AI until I can afford a faster server PC.
Thanks so much for all your help on this issue.

[timothepoznanski]

I have the same kind of machine actually. 😊 Only small models run on it, and the answers are slow and often not very accurate. I also tested with the Anthropic API: it works really well, but your notes' content is sent to Anthropic, so that was just for testing. In the end I keep qwen on a local Ollama just to validate that the feature works, but for real daily use it's not powerful enough. So your workstation option (or a cloud provider if you're okay with the privacy trade-off) is probably the way to go.