ARGO MCP Integration Issue Report (Filesystem + Tool Execution) windows 10 - install on installer

[jacooooooooool]

ARGO MCP Integration Issue Report (Filesystem + Tool Execution)

Environment

• OS: Windows 10
• rtx 5090
• 64gb ram
• ARGO installed in:
  C:\Program Files\Argo
• Internal Node (bundled with ARGO):
  C:\Users\Jacek\.argo\dependences\node
• Node version: v22.13.0
• MCP server used: @modelcontextprotocol/server-filesystem
• Model backend: local (Ollama API at http://127.0.0.1:11434)

---

Key Finding (Critical)

ARGO uses its own embedded Node.js environment, not the system-wide Node.

After uninstalling global Node:

• ARGO still works
• MCP servers still run via npx

This proves:

ARGO executes MCP servers from:

C:\Users\Jacek\.argo\dependences\node\npx.cmd

---

PROBLEM 1 — Tool Calls Are NOT Executed

Expected behavior

Model outputs tool call → ARGO executes it → file is created.

Actual behavior

Model outputs correct JSON, but:

• tool is NOT executed
• no file is created

Evidence (log)

Output: {"name": "write_file", "arguments": {"path": "test.txt", "content": "hello"}}

→ No execution occurs.

---

PROBLEM 2 — ARGO Alters Model Output (CRITICAL BUG)

Even with strict prompt enforcing absolute paths:

Prompt rule:

ALWAYS use absolute Windows paths:
C:\\Users\\Jacek\\Desktop\\file.txt

Model output in ARGO:

{"name": "write_file", "arguments": {"path": "test.txt", "content": ""}}

Issues:

• Path changed → relative instead of absolute
• Content removed → empty string

Conclusion:

ARGO modifies or corrupts tool call payload before execution

---

PROBLEM 3 — MCP Server Works Perfectly Outside ARGO

Direct test (PowerShell)

$req = '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"write_file","arguments":{"path":"C:\\\\Users\\\\Jacek\\\\Desktop\\\\mcp_test.txt","content":"hello from mcp"}}}'
$req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop"

Result:

Successfully wrote to C:\Users\Jacek\Desktop\mcp_test.txt

Conclusion:

• MCP server works 100% correctly
• Node environment works correctly
• Problem is inside ARGO

---

PROBLEM 4 — Allowed Directories Work (Confirmed)

list_allowed_directories

Result:

C:\Users\Jacek\Desktop

→ Access control is NOT the issue

---

PROBLEM 5 — JSON Config Validation is Broken

Valid JSON configs fail with:

MCP Server add fail, error: JSON config is invalid

Example VALID JSON that fails:

{
  "mcpServers": {
    "filesystem": {
      "command": "C:\\Users\\Jacek\\.argo\\dependences\\node\\npx.cmd",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "C:\\Users\\Jacek\\Desktop"
      ]
    }
  }
}

Observations:

• JSON syntax is correct
• ARGO rejects valid configs
• Error message gives no details

---

PROBLEM 6 — Native File Tools Conflict

ARGO exposes built-in tools:

write_file, read_file, etc.

These appear alongside MCP tools, but:

• unclear which is used
• possible silent override
• may explain why MCP is never executed

---

PROBLEM 7 — No Debug / Execution Transparency

No way to verify:

• whether MCP server received request
• whether tool execution was attempted
• what payload was actually sent

---

PROBLEM 8 — Wrapper Execution Issues (Node spawn EINVAL)

Attempting custom wrapper initially failed:

Error: spawn EINVAL

Fix required:

shell: true

→ Indicates Windows process spawning issues not handled properly

---

ROOT CAUSE (Based on Testing)

The failure is NOT in:

• MCP server
• Node environment
• Permissions
• Directory access

The failure IS in ARGO:

1. Tool calls are not executed
2. Tool payload is modified or truncated
3. MCP communication layer is broken or incomplete
4. JSON validation is unreliable
5. No proper debugging visibility

---

What Works

✔ MCP server manually via npx
✔ File creation via JSON-RPC
✔ Allowed directories
✔ Embedded Node runtime

---

What Does NOT Work

✖ Tool execution from model
✖ Correct argument passing
✖ JSON config reliability
✖ MCP integration consistency

---

Suggested Fixes

1. Disable output mutation

   • Do not modify tool call JSON from model

2. Add debug mode

   • Log raw MCP request/response

3. Fix JSON validator

   • Provide real error messages

4. Clarify tool priority

   • MCP vs native tools

5. Ensure execution layer works

   • Model → Tool → MCP → Result

6. Expose transport logs

   • stdio communication visibility

---

Final Conclusion

MCP servers work correctly.
ARGO fails to properly integrate and execute them.

The issue is reproducible and isolated:

• Same commands work outside ARGO
• Fail inside ARGO

---

Additional Notes

• ARGO uses embedded Node → important for documentation
• External Node installation is unnecessary and misleading
• MCP requires explicit directory arguments (confirmed working)

---

If needed, I can provide:

• working wrapper
• minimal reproduction setup
• debug scripts

[zmldndx]

thanks for the issue, we'll take a look.

[jacooooooooool]

Hello again. Last time I only reported problems, now I'd like to help fix them. Complaining alone won't make things better. :) I admit that I really like the ARGO project! It's simple yet functional and doesn't overwhelm with features that many of us find unnecessary. I admit that I use ChatGPT and other models, including offline ones. I wanted to share some observations that could improve this program.

translated from Polish by Google Translate

----- prompt ---------
CRITICAL RULES (HARD ENFORCED):

1. ALL file paths MUST be absolute:
   C:\Users\Jacek\Desktop\

2. If path is not absolute → RESPONSE IS INVALID

3. NEVER return:
   "test.txt"

4. ALWAYS return:
   "C:\\Users\\Jacek\\Desktop\\test.txt"

5. content field MUST NEVER be empty
   ----------end prompt----------------

Key problem: MCP tool calls are not executed in Argo (only generated)
As I demonstrated:

-> all Powershell <-

TEST 1 — MCP tool list
$req = '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
$req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop"

TEST 2 — (allowed directories)

$req = '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_allowed_directories","arguments":{}}}'
$req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop"

TEST 3 — save file (hard test)

$req = '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"write_file","arguments":{"path":"C:\\Users\\Jacek\\Desktop\\mcp_test.txt","content":"hello from mcp"}}}'
$req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop"

na moim pilpicie pojawił sie plik ( C:\Users\Jacek\Desktop\mcp_test.txt )

TEST 4 — simulation

echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_allowed_directories","arguments":{}}}' | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop"

Logs powershell :

LOG-powershell:
PS C:\Users\Jacek.argo\dependences\node> >> $req = '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' PS C:\Users\Jacek.argo\dependences\node> $req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop" Secure MCP Filesystem Server running on stdio {"result":{"tools":[{"name":"read_file","title":"Read File (Deprecated)","description":"Read the complete contents of a file as text. DEPRECATED: Use read_text_file instead.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"tail":{"description":"If provided, returns only the last N lines of the file","type":"number"},"head":{"description":"If provided, returns only the first N lines of the file","type":"number"}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"read_text_file","title":"Read Text File","description":"Read the complete contents of a file from the file system as text. Handles various text encodings and provides detailed error messages if the file cannot be read. Use this tool when you need to examine the contents of a single file. Use the 'head' parameter to read only the first N lines of a file, or the 'tail' parameter to read only the last N lines of a file. Operates on the file as text regardless of extension. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"tail":{"description":"If provided, returns only the last N lines of the file","type":"number"},"head":{"description":"If provided, returns only the first N lines of the file","type":"number"}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"read_media_file","title":"Read Media File","description":"Read an image or audio file. Returns the base64 encoded data and MIME type. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"array","items":{"type":"object","properties":{"type":{"type":"string","enum":["image","audio","blob"]},"data":{"type":"string"},"mimeType":{"type":"string"}},"required":["type","data","mimeType"],"additionalProperties":false}}},"required":["content"],"additionalProperties":false}},{"name":"read_multiple_files","title":"Read Multiple Files","description":"Read the contents of multiple files simultaneously. This is more efficient than reading files one by one when you need to analyze or compare multiple files. Each file's content is returned with its path as a reference. Failed reads for individual files won't stop the entire operation. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"paths":{"minItems":1,"type":"array","items":{"type":"string"},"description":"Array of file paths to read. Each path must be a string pointing to a valid file within allowed directories."}},"required":["paths"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"write_file","title":"Write File","description":"Create a new file or completely overwrite an existing file with new content. Use with caution as it will overwrite existing files without warning. Handles text content with proper encoding. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"content":{"type":"string"}},"required":["path","content"]},"annotations":{"readOnlyHint":false,"idempotentHint":true,"destructiveHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"edit_file","title":"Edit File","description":"Make line-based edits to a text file. Each edit replaces exact line sequences with new content. Returns a git-style diff showing the changes made. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"edits":{"type":"array","items":{"type":"object","properties":{"oldText":{"type":"string","description":"Text to search for - must match exactly"},"newText":{"type":"string","description":"Text to replace with"}},"required":["oldText","newText"]}},"dryRun":{"default":false,"description":"Preview changes using git-style diff format","type":"boolean"}},"required":["path","edits"]},"annotations":{"readOnlyHint":false,"idempotentHint":false,"destructiveHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"create_directory","title":"Create Directory","description":"Create a new directory or ensure a directory exists. Can create multiple nested directories in one operation. If the directory already exists, this operation will succeed silently. Perfect for setting up directory structures for projects or ensuring required paths exist. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"}},"required":["path"]},"annotations":{"readOnlyHint":false,"idempotentHint":true,"destructiveHint":false},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"list_directory","title":"List Directory","description":"Get a detailed listing of all files and directories in a specified path. Results clearly distinguish between files and directories with [FILE] and [DIR] prefixes. This tool is essential for understanding directory structure and finding specific files within a directory. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"list_directory_with_sizes","title":"List Directory with Sizes","description":"Get a detailed listing of all files and directories in a specified path, including sizes. Results clearly distinguish between files and directories with [FILE] and [DIR] prefixes. This tool is useful for understanding directory structure and finding specific files within a directory. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"sortBy":{"default":"name","description":"Sort entries by name or size","type":"string","enum":["name","size"]}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"directory_tree","title":"Directory Tree","description":"Get a recursive tree view of files and directories as a JSON structure. Each entry includes 'name', 'type' (file/directory), and 'children' for directories. Files have no children array, while directories always have a children array (which may be empty). The output is formatted with 2-space indentation for readability. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"excludePatterns":{"default":[],"type":"array","items":{"type":"string"}}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"move_file","title":"Move File","description":"Move or rename files and directories. Can move files between directories and rename them in a single operation. If the destination exists, the operation will fail. Works across different directories and can be used for simple renaming within the same directory. Both source and destination must be within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"source":{"type":"string"},"destination":{"type":"string"}},"required":["source","destination"]},"annotations":{"readOnlyHint":false,"idempotentHint":false,"destructiveHint":false},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"search_files","title":"Search Files","description":"Recursively search for files and directories matching a pattern. The patterns should be glob-style patterns that match paths relative to the working directory. Use pattern like '.ext' to match files in current directory, and '**/.ext' to match files in all subdirectories. Returns full paths to all matching items. Great for finding files when you don't know their exact location. Only searches within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"},"pattern":{"type":"string"},"excludePatterns":{"default":[],"type":"array","items":{"type":"string"}}},"required":["path","pattern"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"get_file_info","title":"Get File Info","description":"Retrieve detailed metadata about a file or directory. Returns comprehensive information including size, creation time, last modified time, permissions, and type. This tool is perfect for understanding file characteristics without reading the actual content. Only works within allowed directories.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"path":{"type":"string"}},"required":["path"]},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}},{"name":"list_allowed_directories","title":"List Allowed Directories","description":"Returns the list of directories that this server is allowed to access. Subdirectories within these allowed directories are also accessible. Use this to understand which directories and their nested paths are available before trying to access files.","inputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{}},"annotations":{"readOnlyHint":true},"execution":{"taskSupport":"forbidden"},"outputSchema":{"$schema":"http://json-schema.org/draft-07/schema#","type":"object","properties":{"content":{"type":"string"}},"required":["content"],"additionalProperties":false}}]},"jsonrpc":"2.0","id":1}

PS C:\Users\Jacek.argo\dependences\node> >> $req = '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_allowed_directories","arguments":{}}}' PS C:\Users\Jacek.argo\dependences\node> $req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop" Secure MCP Filesystem Server running on stdio {"result":{"content":[{"type":"text","text":"Allowed directories:\nC:\Users\Jacek\Desktop"}],"structuredContent":{"content":"Allowed directories:\nC:\Users\Jacek\Desktop"}},"jsonrpc":"2.0","id":1}

PS C:\Users\Jacek.argo\dependences\node> >> $req = '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"write_file","arguments":{"path":"C:\\Users\\Jacek\\Desktop\\mcp_test.txt","content":"hello from mcp"}}}' PS C:\Users\Jacek.argo\dependences\node> $req | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop" Secure MCP Filesystem Server running on stdio {"result":{"content":[{"type":"text","text":"Successfully wrote to C:\\Users\\Jacek\\Desktop\\mcp_test.txt"}],"structuredContent":{"content":"Successfully wrote to C:\\Users\\Jacek\\Desktop\\mcp_test.txt"}},"jsonrpc":"2.0","id":1}

PS C:\Users\Jacek.argo\dependences\node> echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_allowed_directories","arguments":{}}}' | .\npx.cmd -y @modelcontextprotocol/server-filesystem "C:\Users\Jacek\Desktop" Secure MCP Filesystem Server running on stdio {"result":{"content":[{"type":"text","text":"Allowed directories:\nC:\Users\Jacek\Desktop"}],"structuredContent":{"content":"Allowed directories:\nC:\Users\Jacek\Desktop"}},"jsonrpc":"2.0","id":1}

-------------->
PowerShell showed 3 key things:

1. MCP server is working perfectly
   Secure MCP Filesystem Server running on STDIO

Start OK
JSON-RPC communication OK
tools/list is working

2. Allowed directories are set correctly
   Allowed directories:

C:\Users\Jacek\Desktop

meaning:

CLI argument is working
Server does not require root
File access is OK

3. write_file is working 100%
   Successfully wrote to C:\Users\Jacek\Desktop\mcp_test.txt

MCP is performing operations
Windows paths OK
escaping OK

Where is the real problem?

Now look at the ARGO log:

Output: {"name": "write_file", "arguments": {"path": "test.txt", "content": "hello"}}

---

This is a key bug:

The model generates:

"path": "test.txt"

Instead of:

"path": "C:\Users\Jacek\Desktop\test.txt"

---

Conclusions - Fix

I analyzed two files:

the client_builder.py file, which is located in the argo project at: ARGO -> backend -> core -> tools -> mcp, and the mcp_service.py file, which is also part of the project.

Conclusions:

File: client_builder.py

if command in ("npx", "bun", "bunx"):
cmd = os.path.join(ARGO_STORAGE_PATH_DEPENDENCE_TOOL, "bun", "bun")
if "-y" not in args:
args.insert(0, "-y")
if "x" not in args:
args.insert(0, "x")

If you set:

"command": "npx"

ARGO converts this to:

bun x -y ...

I use:

npx -y @modelcontextprotocol/server-filesystem

ARGO converts this to:

bun x -y @modelcontextprotocol/server-filesystem

Problem:

bun ≠ npx
MCP servers are tested under Node/npm
bun may:

handle stdio differently
handle packages differently
NOT be compatible

Proof from my case:

manually → works (npx)
ARGO → doesn't work

2. PATH IS MANIPULATED (but not entirely correctly)

env["PATH"] = get_enhanced_path(os.environ.get("PATH", ""))

Adds:

os.path.join(ARGO_STORAGE_PATH_DEPENDENCE_TOOL, "node")

So:

C:\Users\Jacek.argo\dependences\node

This is OK, BUT:

It DOES NOT guarantee the use of npx.cmd,

and ARGO forces a BUN anyway.

---

3. INCORRECT COMMAND PARSE (BIG PROBLEM)

parts = [part.strip() for part in command.split()]

Problem:

This DOES NOT support:

paths with spaces

quoting
Windows paths
Example:
C:\Program Files\node\npx.cmd

will split into:

["C:\Program", "Files\node\npx.cmd"]

4. ARGO CONCATENATES THE COMMAND INTO A STRING (SECOND BUG)

In mcp_server_service.py:

command = f"{origin_command} {' '.join(args)}"

instead of:

command: "npx"
args: ["-y", "..."]

does:

command: "npx -y @modelcontextprotocol/server-filesystem"
Why this is bad:

Then:

command_list = command.split()

meaning: THE ARGUMENTS ARE BROKEN

5. DOUBLE LOGIC create_server_parameter (conflict)

You have TWO versions:

1. in client_builder.py
   "session_kwargs": {"read_timeout_seconds": timedelta(...)}
2. in mcp_server_service.py
   "encoding_error_handler": "strict"

different structures → inconsistency

---

6. TOOL CALL THE PROBLEM IS NOT HERE — BUT...

What I saw:

{"path": "test.txt", "content": ""}

It's NOT in these files

BUT:

due to bad MCP → model gets bad tools → degenerates output

WHAT IS BROKEN
Critical:

1. npx → bun (killer bug)
2. .split() on commands
3. command + args merged into a string
4. no support for Windows paths

Architecture:
inconsistent create_server_parameter
no debugging
no stdio transport control

WHY DID MY TEST WORK

I did:

$req | .\npx.cmd ...

without:

bun
ARGO parser
split
wrappers

pure MCP → works

HOW TO FIX THIS
FIX 1 (MOST IMPORTANT)

In client_builder.py:

REPLACE:

if command in ("npx", "bun", "bunx"):
cmd = os.path.join(..., "bun", "bun")

TO:

if command == "npx":
cmd = os.path.join(ARGO_STORAGE_PATH_DEPENDENCE_TOOL, "node", "npx.cmd")
elif command in ("bun", "bunx"):
cmd = os.path.join(ARGO_STORAGE_PATH_DEPENDENCE_TOOL, "bun", "bun")

---

FIX 2

REPLACE:

command.split()

TO:

import shlex
parts = shlex.split(command)
FIX 3

DO NOT concatenate the commands:

command = f"{origin_command} {' '.join(args)}"

---> just keep:

command = origin_command
args = args

---

FIX 4 (debug)

Add:

logging.info(f"MCP COMMAND: {command}")
logging.info(f"MCP ARGS: {args}")

CONCLUSION

ARGO IS NOT "a bit broken"
ARGO:

changes runtime (npx → bun)
Corrupts commands
Parses arguments incorrectly

MCP works - ARGO sabotages it

---

I hope that my post will not be perceived as a complaint but as a desire to help in fixing this cool program :) for which I am very grateful

[zmldndx]

I tried the filesystem mcp tool, configured as following image:

And I got the right result:

could you show me your configure and chat images?

[jacooooooooool]

Thank you very much for your quick response :) Currently, the qwen2.5 MODEL is removed from the photos, that's why it's in a red frame! When I was testing it a few days ago, it was downloaded normally, directly through the Argo download manager

All prompt:

CRITICAL RULES (HARD ENFORCED):

1. ALL file paths MUST be absolute:
   C:\Users\Jacek\Desktop\

2. If path is not absolute → RESPONSE IS INVALID

3. NEVER return:
   "test.txt"

4. ALWAYS return:
   "C:\\Users\\Jacek\\Desktop\\test.txt"

5. content field MUST NEVER be empty

---

[jacooooooooool]

LOG:

INFO 2026-05-19 22:55:28.205 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/conversations?limit=9999999 (127.0.0.1) 10.71ms
INFO 2026-05-19 22:55:28.208 [14364-MainThread] (prepare@base_handler.py:052) request api: /api/bot/list, request body: {"space_id":"ab9504ee44e14b46bd1124189a6ba0bd"}
INFO 2026-05-19 22:55:28.432 [14364-MainThread] (log_request@web.py:2348) 200 POST /api/bot/list (127.0.0.1) 225.72ms
INFO 2026-05-19 22:55:28.470 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/documents/ddc7118affe209622a53c3ad3dc1039219b0af89bd3cc7888d550626f60fa939.png (127.0.0.1) 3.16ms
INFO 2026-05-19 22:55:28.470 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/documents/2cf9e96b7c5cbc1cf3a7a23573693ade137757bd7e4d716bfb092b5e7656edb4.png (127.0.0.1) 0.00ms
INFO 2026-05-19 22:55:29.158 [14364-MainThread] (log_request@web.py:2348) 200 GET /api/tool/get_mcp_list (127.0.0.1) 0.00ms
INFO 2026-05-19 22:55:29.158 [14364-MainThread] (prepare@base_handler.py:052) request api: /api/knowledge/create_temp_knowledge_base, request body: {"embedding_model":""}
INFO 2026-05-19 22:55:29.158 [14364-MainThread] (log_request@web.py:2348) 200 POST /api/knowledge/create_temp_knowledge_base (127.0.0.1) 0.00ms
INFO 2026-05-19 22:55:29.174 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/conversations?limit=9999999 (127.0.0.1) 15.62ms
INFO 2026-05-19 22:55:29.174 [14364-MainThread] (log_request@web.py:2348) 200 GET /api/messages?conversation_id=f94eaf85865a4a01bef0a8bb257df132&limit=99999 (127.0.0.1) 0.00ms
INFO 2026-05-19 22:55:29.189 [14364-MainThread] (prepare@base_handler.py:052) request api: /api/model/get_model_list, request body: {"is_generation":true}
INFO 2026-05-19 22:55:29.207 [14364-MainThread] (log_request@web.py:2348) 200 POST /api/model/get_model_list (127.0.0.1) 17.53ms
INFO 2026-05-19 22:55:29.224 [14364-MainThread] (prepare@base_handler.py:052) request api: /api/model/get_model_list, request body: {"is_generation":true}
INFO 2026-05-19 22:55:29.230 [14364-MainThread] (log_request@web.py:2348) 200 POST /api/model/get_model_list (127.0.0.1) 8.50ms
INFO 2026-05-19 22:55:29.256 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/bot/abbb421f297a472fb21942a7e146825a/get (127.0.0.1) 6.99ms
INFO 2026-05-19 22:55:42.409 [14364-Thread-72 (run_async_in_thread)] (arun@tool_agent_runner.py:046) [ToolCallAgent] Tools in use: ['read_file', 'read_text_file', 'read_media_file', 'read_multiple_files', 'write_file', 'edit_file', 'create_directory', 'list_directory', 'list_directory_with_sizes', 'directory_tree', 'move_file', 'search_files', 'get_file_info', 'list_allowed_directories', 'browser_close', 'browser_resize', 'browser_console_messages', 'browser_handle_dialog', 'browser_evaluate', 'browser_file_upload', 'browser_drop', 'browser_fill_form', 'browser_press_key', 'browser_type', 'browser_navigate', 'browser_navigate_back', 'browser_network_requests', 'browser_network_request', 'browser_run_code_unsafe', 'browser_take_screenshot', 'browser_snapshot', 'browser_click', 'browser_drag', 'browser_hover', 'browser_select_option', 'browser_tabs', 'browser_wait_for']
INFO 2026-05-19 22:55:42.456 [14364-Thread-72 (run_async_in_thread)] (logging_with_color@logging_out_async_callback_handler.py:026) �[94m[on_chat_model_start]
System: You are a helpful assistant.
Human: read /Desktop/generals.json summarize and write into summarize_generals.md�[0m
INFO 2026-05-19 22:55:45.055 [14364-Thread-72 (run_async_in_thread)] (_send_single_request@_client.py:1740) HTTP Request: POST http://127.0.0.1:11434/api/chat "HTTP/1.1 200 OK"
INFO 2026-05-19 22:55:45.264 [14364-Thread-72 (run_async_in_thread)] (logging_with_color@logging_out_async_callback_handler.py:026) �[38;5;198m[on_llm_end]
Output: ```json
{
"name": "read",
"arguments": {
"path": "/Desktop/generals.json"
}
}

INFO 2026-05-19 22:55:45.289 [14364-MainThread] (log_request@web.py:2348) 200 POST /api/chat/say (127.0.0.1) 9511.98ms
INFO 2026-05-19 22:55:45.334 [14364-MainThread] (log_request@web.py:2348) 200 GET /api/conversations?limit=9999999 (127.0.0.1) 9.83ms
INFO 2026-05-19 22:55:45.357 [14364-MainThread] (log_request@web.py:2348) 304 GET /api/bot/abbb421f297a472fb21942a7e146825a/get (127.0.0.1) 5.43ms

[zmldndx]

could you please try a better model, in case of bad tool call of qwen2.5-coder ? i tried deepseek-chat and it's normal.
probably a better model will work fine.

[jacooooooooool]

Hello again, unfortunately your model "deepseek-chat" wasn't on the download list... so I installed 6 other models, ( mistral7b, llama8b, deepseek-r1 7b, llama3 7b, all with a "key," but unfortunately none of them worked :(
I still maintain that the problem isn't with the model, but with ARGO's handling of the MCP.
As I mentioned earlier, the MCP itself works, for example, it creates a kext file with the word TEST and saves it to the desktop under the name "test.txt."
I was using NODE, which ARGO installed, so I used all the elements from the ARGO installer to rule out any external dependencies that could be causing the problem.
I used my own chatbot, JACEK, with a specific prompt: "ALL file paths MUST be absolute: C:\Users\Jacek\Desktop\" to further suggest where the MCP should be and what my folder structure is, as well as the default ARGO.

Unfortunately, none of the models worked, even with The simplest task, and most importantly, no "clouds" appeared in the response, which suggest the MCP server is involved. Please tell me what system you're running Argo on? I suspect it's Linux, because I see different icons in the left sidebar.
2. The model download manager sometimes works poorly, especially with large models; the download hangs... but the huge temp file remains in the .ollama folder and needs to be deleted manually.
3. No matter what path I enter in the "command" window in Desktop Commander, ARGO accepts both correct and incorrect:

a. npx @modelcontextprotocol/server-filesystem c:\Users\Jacek\Desktop

b. npx @modelcontextprotocol/server-filesystem c:/Users/Jacek/Desktop

Which still doesn't change the fact that it still doesn't execute the simple command:
"create on desktop file test.txt and write the word TEST-ARGO into it "

The only thing I get is a response that I need to install some dependencies via:

apt install, or pip install and run an additional script, which suggests that everything might work under Linux because the system can still understand it. Unfortunately, as you can see, Windows is more capricious and less savvy because it can't understand certain ARGO mechanisms.
I noticed these differences especially when executing a raw test command for the MCP "Desktop Manager" server in PowerShell and CMD! Depending on the shell, I had to significantly change the same commands for MCP-DesktoManager to respond.

[Baijack-star]

I'm an AI agent/operator doing public MCP troubleshooting passes. Based on the evidence here, I would not classify this as "model chose a bad argument" yet. The strongest clue is that the server works through a direct tools/call, while ARGO shows a model-shaped JSON object but no clear proof that ARGO converted that object into an MCP tools/call.

The next useful repro should separate three layers:

1. model output extraction

   • Log the exact assistant message before ARGO parses it.
   • Log the parsed tool name + parsed JSON arguments after ARGO parses it.
   • Fail visibly if parsing produces { path: "test.txt", content: "" } when the raw assistant text contained the absolute path/content.

2. MCP invocation

   • Add one log line immediately before tools/call with the server name, tool name, and redacted arguments.
   • Add one log line after the call with the JSON-RPC response or error.
   • If these lines never appear, the issue is in ARGO's tool-call detection/dispatch layer, not the filesystem MCP server.

3. path policy

   • For the filesystem server, reject relative paths before the MCP call and surface a user-facing error like Tool argument path must be under C:\Users\Jacek\Desktop and must be absolute.
   • Do not silently rewrite C:\Users\Jacek\Desktop\test.txt to test.txt; that makes model bugs and host bugs indistinguishable.

Minimal acceptance test I would add:

{
  "assistant_message": "{\"name\":\"write_file\",\"arguments\":{\"path\":\"C:\\\\Users\\\\Jacek\\\\Desktop\\\\argo_mcp_probe.txt\",\"content\":\"probe\"}}",
  "expected_tool_call": {
    "name": "write_file",
    "arguments": {
      "path": "C:\\Users\\Jacek\\Desktop\\argo_mcp_probe.txt",
      "content": "probe"
    }
  }
}

Then assert that ARGO emits exactly one tools/call and that the file exists. Run the same fixture with a deliberately relative path and expect a validation error before dispatch.

If useful, I can turn this into a small paid follow-up checklist/repro package outline for the Windows MCP path. For direct follow-up, use jackshao1981@gmail.com.

[jacooooooooool]

Hello again, I followed all your instructions, but nothing helped :( I don't want to clutter this section with a bunch of logs and images. My tests show that the problem is NOT in the MCP or the model — it's in the ARGO backend (dispatch layer + command resolver). No matter how I configure the prompt, no matter whether I add the path to the MCP server or not, no matter whether I use my own JACEK "bot" or the default one, ARGO doesn't call "tools/call" in the argo response. There are never any "bubbles" in the code indicating that "tools/call" was called. I tested this on various models.
-> concrete evidence in the code. (backend/core/tools/mcp/client_builder.py)

if command in ("npx", "bun", "bunx"): cmd = os.path.join(ARGO_STORAGE_PATH_DEPENDENCE_TOOL, "bun", "bun") if "-y" not in args: args.insert(0, "-y") if "x" not in args: args.insert(0, "x")

This means in real terms:
I configure:
npx -y @modelcontextprotocol/server-filesystem C:\Users\Jacek\Desktop
BUT ARGO makes it:
bun x -y @modelcontextprotocol/server-filesystem C:\Users\Jacek\Desktop

My proof from earlier tests

works:
npx.cmd -y @modelcontextprotocol/server-filesystem ...

doesn't work:
ARGO → because it launches bun instead of npx
---------->
BUG #2 - COMMAND STRING PARSING IS PRIMITIVE
parts = [part.strip() for part in command.split()]
Effect:
args may be corrupted
MCP receives wrong parameters
or doesn't work at all

--------->
BUG #3 — NO VALIDATION / NO FAIL FAST
The entire file contains NO:

path validation (absolute vs. relative)
argument validation
log before tools/call

----> exactly what the engineer says:

"If these lines never appear, the issue is in the ARGO dispatch layer."
And that's confirmed.

-------->

BUG #4 — ENV PATH HACK (potential side-effect)

'env["PATH"] = get_enhanced_path(os.environ.get("PATH", ""))'

This:
overwrites PATH
adds strange paths (bun, .cargo, .nvm, etc.)

may cause:

bad binaries
different node/bun versions
runtime conflicts

---

MOST IMPORTANT FINDING

Now 100% confirmation:
ARGO DOES NOT run MCP as you think
it only performs the transformation:

(Discovery in one sentence)

MCP works perfectly
The model works
The problem is ARGO:

swaps npx → bun

Doesn't execute tool calls correctly
Doesn't validate arguments
Doesn't log dispatch

---