• State Suggested Answer
  • Replies 1 reply
  • Answers 1 answer
  • Subscribers 49 subscribers
  • Views 86 views
  • Users 0 members are here
Options
Related

IshRemoteMcpServer with AI Toolkit or Continue extension in VSCode

Colin Ferguson

My request: Can you figure out how to get IshRemoteMcpServer to work correctly with either the AI Toolkit or Continue extensions in VSCode?

Here's the reason: Our corporate policies prevent us from using most external AI models such as github Copilot or Claude. So to test IshRemoteMcpServer with our internal Tridion Docs environments I have to set up a local LLM instance (ollama in my case) and get it running with VSCode. It’s very slow and periodically throws errors (because VSCode throws a fit if it can't see its native Copilot when chatting is occurring even to another LLM) but it (sort of) works… I can start the MCP server and see the ishremote commands in the Tools menu. The model seems to recognize them but absolutely will NOT actually execute them. For example, I tell it to start an ishremote session and it shows me how to write powershell code to do it.

So I thought I'd swap in a more powerful AI with better MCP tools handling. I set up an instance of ollama on a server with a decent GPU, and an LLM (llama2 for now) can now be addressed over port 11434 on that server. The problem now is that there is no way to configure a local LLM server in the native VSCode chat via a URL.

So I tried installing the AI Toolkit extension, which supports local configs like this. Then I configured IshRemoteMcpServer and started the server via the extension interface. The first time, it times out: Unable to connect to MCP Server 'IshRemoteMcpServer'. Error: MCP error -32001: Request timed out. Please verify your mcp.json configuration settings. MCP error -32001: Request timed out. The log file ended with: 

{"Level":"Info","Message":"Starting MCP Server","Timestamp":"2026-01-22T21:36:02.1876440-05:00"}

The second time, I get Unable to connect to MCP Server 'IshRemoteMcpServer'. Error: Server's protocol version is not supported: 0.3.0. Please verify your mcp.json configuration settings. Server's protocol version is not supported: 0.3.0. The log file ended with:

{"Level":"Info","RequestId":1,"Request":"{\"method\":\"initialize\",\"params\":{\"protocolVersion\":\"2025-06-18\",\"capabilities\":{},\"clientInfo\":{\"name\":\"vsc-aitoolkit-IshRemoteMcpServer\",\"version\":\"0.28.2\"}},\"jsonrpc\":\"2.0\",\"id\":1}","Message":"Processing request","Timestamp":"2026-01-22T21:36:02.2585314-05:00"}

I tried the same exercise with the Continue extension and got exactly the same behavior.

Log file for the second case attached. The log file for the first case looks exactly the same, except it ends one line earlier.IshRemoteMcpServer.log
emoji
Rate translation
×
Suggest better translation
×
Parents
  • 0


    For others watching this thread and think what is "ISHRemoteMcpServer", Colin had a look at the prerelease. The work in progress is hiding in plain sight: https://github.com/RWS/ISHRemote/blob/master/Doc/ReleaseNotes-ISHRemote-8.2.md holding some animated gifs for inspiration. This is typically the source of the actual RWS Community blog post. But absolutely feel free to have a look already. The main goal for this first version is to have ISHRemoteMcpServer working in Visual Studio Code with standard solutions like Claude, Gemini, OpenAI, Mistral, etc 

    Now given your question Colin, it seems that your corporate policies put in a lot of limitations for you. As you probably noticed in my description I stepped away from the (less expensive, older) GPT-4.1 to more powerful (more expensive, recent) Claud Sonnet 4.5 because it simply gave me a lot better result. These results for me was its choice on when to execute something, but also its choices in how it approached co-writing a PowerShell script. That virtual assistant behaves differently, some suggests other execute more. That is something that ISHRemoteMcpServer does not really control. So you were right in switching the model 'out' or 'up'. There is no one-size fits all.

    You mention "llama2" which is quite old, and even your suggestion "Continue" on https://docs.continue.dev/guides/ollama-guide mentions to use Llama 3.1 or Mistral or higher with MCP Tools. Can you try other more recent ones?

    Regarding the MCP Protocol Version "0.3.0". I've updated the latest PowerShellGallery-hosted Prerelease (8.2.0-preview259), so you can update and try again. Hopefully the below story unblocks your efforts, and do not hesitate to write down some tips and tricks for people that have similar corporate policies. See https://modelcontextprotocol.io/specification/2025-11-25/basic/lifecycle and in turn Update ISHRemoteMcpServer MCP Protocol Version from '0.3.0' to '2024-11-05' · Issue #228 · RWS/ISHRemote https://github.com/RWS/ISHRemote/issues/228 

    Best wishes,
    Dave
    emoji
Reply
  • 0


    For others watching this thread and think what is "ISHRemoteMcpServer", Colin had a look at the prerelease. The work in progress is hiding in plain sight: https://github.com/RWS/ISHRemote/blob/master/Doc/ReleaseNotes-ISHRemote-8.2.md holding some animated gifs for inspiration. This is typically the source of the actual RWS Community blog post. But absolutely feel free to have a look already. The main goal for this first version is to have ISHRemoteMcpServer working in Visual Studio Code with standard solutions like Claude, Gemini, OpenAI, Mistral, etc 

    Now given your question Colin, it seems that your corporate policies put in a lot of limitations for you. As you probably noticed in my description I stepped away from the (less expensive, older) GPT-4.1 to more powerful (more expensive, recent) Claud Sonnet 4.5 because it simply gave me a lot better result. These results for me was its choice on when to execute something, but also its choices in how it approached co-writing a PowerShell script. That virtual assistant behaves differently, some suggests other execute more. That is something that ISHRemoteMcpServer does not really control. So you were right in switching the model 'out' or 'up'. There is no one-size fits all.

    You mention "llama2" which is quite old, and even your suggestion "Continue" on https://docs.continue.dev/guides/ollama-guide mentions to use Llama 3.1 or Mistral or higher with MCP Tools. Can you try other more recent ones?

    Regarding the MCP Protocol Version "0.3.0". I've updated the latest PowerShellGallery-hosted Prerelease (8.2.0-preview259), so you can update and try again. Hopefully the below story unblocks your efforts, and do not hesitate to write down some tips and tricks for people that have similar corporate policies. See https://modelcontextprotocol.io/specification/2025-11-25/basic/lifecycle and in turn Update ISHRemoteMcpServer MCP Protocol Version from '0.3.0' to '2024-11-05' · Issue #228 · RWS/ISHRemote https://github.com/RWS/ISHRemote/issues/228 

    Best wishes,
    Dave
    emoji
Children
No Data