[QUESTION] Qwen code cli

[Alphapage]

Question

Hello,
I am trying to use arbor with qwen cli, but it shows errors:
✖ MCP ERROR (arbor): Error: SSE error: TypeError: fetch failed: │
│ unknown scheme │
│ ✖ Error during discovery for server 'arbor': SSE error: │
│ TypeError: fetch failed: unknown scheme │
│ ✖ MCP ERROR (arbor): Error: SSE error: TypeError: fetch failed: │
│ unknown scheme

My settings.json is:
{
"mcpServers": {
"arbor": {
"mode": "ws",
"url": "ws://127.0.0.1:8080",
"timeout": 15000
}
},
"$version": 2
}

The cli says that server is running 👍
I do not know if qwen cli an use stdio as bridge with arbor.

Environment

• OS:Debian 12
• Arbor Version: latest

It seems arbor is indexing a lot of unnecessary files: for example, for a flutter project it should only load .dart files, but it loads some other extensions if they are located in subfolders. Is there a way to filter the files for indexing?

Thank you in advance for your help.

[holg]

Hi!

Issue 1: MCP Connection Error

Arbor's MCP server uses stdio transport (standard input/output), not WebSocket or SSE. The ws:// URL in your config won't work because MCP clients need to spawn arbor bridge as a subprocess and communicate via stdin/stdout.

Correct MCP configuration (for stdio-based clients like Claude Desktop):

{
  "mcpServers": {
    "arbor": {
      "command": "arbor",
      "args": ["bridge"]
    }
  }
}

The WebSocket servers (ports 7433 and 8081) are for the visualizer and internal RPC queries, not for MCP clients.

If Qwen CLI only supports WebSocket/SSE transport and not stdio (AFAIK it does), it won't be able to connect to Arbor directly. You'd need to check Qwen CLI's documentation to see if it supports stdio-based MCP servers.

---

Issue 2: Indexing Files

Arbor only indexes these extensions:

• Dart: .dart
• TypeScript/JavaScript: .ts, .tsx, .js, .jsx, .mts, .cts, .mjs, .cjs
• Rust: .rs
• Python: .py, .pyi
• Go: .go
• Java: .java
• C/C++: .c, .h, .cpp, .hpp, .cc, .hh, .cxx, .hxx

For a Flutter project, only .dart files should be indexed. Arbor also respects .gitignore, so files in build/, .dart_tool/, etc. should be skipped.

Could you share which specific file extensions or folders you're seeing indexed that shouldn't be? That would help us identify if there's a bug.

[Alphapage]

From qwen:
"
It seems that the Arbor tools are correctly installed and working, but they're not
detecting the relationships between the code elements in this specific codebase.
This could be due to the nature of Flutter/Dart code structure where the
relationships might not be picked up in the same way as traditional imperative
code.

However, I've successfully demonstrated that:

1. The Arbor tools are available and working in this environment
2. The codebase has been properly indexed with Dart support
3. Basic queries work (finding classes and functions)
4. The refactor and impact analysis tools are operational

The reason the impact analysis shows 0 nodes might be because in Flutter widgets,
the relationships are often expressed through the widget composition tree rather
than direct function calls, which might not be captured in the same way by the
graph analysis.

Let me summarize what I've learned about using Arbor with this codebase:

Summary of Arbor Usage with the Flappy Bird Codebase

I've successfully demonstrated the following:

1. Setup: Updated the Arbor configuration to include Dart language support
2. Indexing: Successfully indexed the entire codebase with 116 files and 1,578
   nodes
3. Basic Queries: Can find classes and functions
4. Analysis Tools: The refactor, explain, and other tools are operational

While the impact analysis shows 0 nodes for the specific classes and functions I
tested, this is likely due to the Flutter widget architecture where relationships
are expressed through composition rather than direct function calls. Arbor is
still working correctly and can analyze this codebase effectively for other types
of queries.

The Arbor system is a powerful code analysis tool that creates a graph
representation of the codebase, allowing for sophisticated analysis of code
relationships, dependencies, and impact of changes.
"

I asked about filtering only .dart because I do not think my project counts 116 files, but maybe I am wrong and I do not know how to list the files arbor has indexed.

[Anandb71]

Thanks for the detailed report, this helps a lot.

To clarify officially:

MCP Transport

Arbor’s MCP server currently supports stdio only (spawned subprocess via stdin/stdout).
The WebSocket servers (ports 7433 / 8081) are not MCP — they are used internally and by the visualizer.

If a client (like Qwen CLI) only supports SSE or WebSocket-based MCP, it unfortunately cannot connect to Arbor at this time.

This is not a configuration issue on your side — it’s a transport mismatch.

File Indexing

Arbor only indexes language-specific source files (e.g. .dart for Flutter).
It also respects .gitignore.

If you’re seeing more files than expected:

• they may be Dart files you didn’t account for
• or generated files not covered by .gitignore

At the moment, there is no CLI command to list exactly which files were indexed.
That’s a good candidate for a future arbor status --files or debug flag.

Flutter & Impact Analysis

Your observation is correct:
Flutter widget relationships are often expressed via composition, not direct calls.
Today, impact analysis focuses on explicit call/import edges, not widget trees.

That’s expected behavior, not a failure.

I’ll leave this issue open for documentation improvements and potential tooling clarity.
Thanks again for the detailed investigation.

[MonzerMahmoud]

Arbor only indexes language-specific source files (e.g. .dart for Flutter).
It also respects .gitignore.

thats explains why adding Platform directories to .arbor/config.json ignore list does not change anything, although setting languages to only dart it stills prints a lot of warnings about linux and windows files.

[Anandb71]

You're right, thanks for pointing this out. Arbor only indexes language-specific files like .dart, but it still walks the folder tree, so platform folders get discovered even though their files are not actually indexed. That’s why you see warnings even when they don’t affect the graph.

The ignore list in .arbor/config.json currently does not stop directory traversal, it only affects indexing. This is why platform directories still show warnings but Dart indexing stays correct.

This will be improved soon. Planned update: silence warnings for non-indexable files and add proper .arborignore support so you can exclude entire folders cleanly. Also adding arbor status --files so you can see exactly which files were indexed.

Appreciate you raising this, it helps smooth out these rough edges.