Fullstack vs Neovim MCP

[ckangnz]

First of all, I'm loving codecompanion and I believe this tool helped the entire neovim community not miss out on AI agent that every other nowadays IDEs have. Thank you so much for a wonderful tool!

Ever since i configured MCP, i have been using Neovim MCP (the default one that comes with MCPHub.nvim) mainly because

• it shows you what commands it's running to find, run, query etc. its doing
• i think its diff is much easier to follow than the fullstack_dev / insert_edit_files tool

But i have been experiencing a bit of issue with MCP at work because the npm server at work goes through vpn, and #{neovim} is usually not ready to be used until all other mcps are loaded. So I'm trying to get used to the default fullstack tool but I cant seem to find a setting where it allows the tool to show me what commands it ran to do what, and what the outcomes were in the chat buffer.
How do I get this behaviour?

[pixlmint]

I believe the CodeCompanionToolStarted/Finished autocmd events might be what you're looking for (https://codecompanion.olimorris.dev/usage/events)

[bassamsdata]

Hello @ckangnz,

• i think its diff is much easier to follow than the fullstack_dev / insert_edit_files tool

I’m interested in what you mean here, could you expand on the diff experience? I tried the Neovim MCP diff myself, but the virtual text and colors made it feel a bit hard to read.

I think a comparable experience would be switching the diff layout to “buffer” instead of the default “float.”

-- https://codecompanion.olimorris.dev/configuration/chat-buffer#diff
require("codecompanion").setup({
  display = {
    diff = {
      provider_opts = {
        inline = {
          layout = "buffer", -- float|buffer - Where to display the diff
        },
      },
    },
  },
})

[ckangnz]

Here's an example of the similar task I've requested on both @{neovim}(TOP) and @{full_stack_dev} (BOTTOM). Note that, @{full_stack_dev} actually did a better job without needing me to tell me which file i'm referring to. I had to add #{buffer} to get the similar response.

Anyways, as you can see @{neovim} shows exactly what the AI is doing to achieve what I have asked for.

**`neovim__read_file` Tool**: Returned the following:

**`neovim__find_files` Tool**: Returned the following:

**`neovim__execute_command` Tool**: Returned the following:

**`neovim__execute_command` Tool**: Returned the following:

and it even shows me the details of the commands it ran to achieve the outcome. This is extremely useful, when your AI is doing curl/grep stuff to automatically do something, and you can learn from what its doing.

Once accepting the file change, it says it ran **neovim__edit_file Tool**: Returned the following: with the change it made on the chat buffer.

Although @{full_stack_dev} has cleaner chat buffer (img below), I have no clue what the agent has done for me (especially when I choose accept all. (I know you can use gD to look at super diff, but i can't tell which prompt made the change).

[ckangnz]

Also i find Neovim MCP easier to 'apply all' because it only applies for the diff of the current prompt. I can just go through the diffs first and accept all when I'm happy.
Once the prompt is completed, it asks me to accept changes again on my next prompt.

The fullstack's 'always accept' is allowing edit for current chat session. Once i select 'always accept', i believe it wont ask for edit permission again for the current chat session? It gets out of control because

1. The chat buffer doesnt show me enough info on what's being changed.
2. Changes are made without showing me the diff.
3. When I open Super Diff, i can't tell which prompt made which diffs.

Hope my explanation makes sense. 🙂

[bassamsdata]

Thank you for all the details! I’ll review them over the weekend and also take a look at the @neovim edit and diff.

When I open Super Diff, I can't tell which prompt made which diffs.

I agree, though it’s not an essential feature I think. That said, with the new super_diff (after v18) and if the PR is merged, it should be possible to track which prompt made which diffs with proper checkpoints.