DEV Community

Cover image for Reading a remote log with Claude Desktop and remotify-mcp
EZsTGB
EZsTGB

Posted on Edited on Originally published at ezstgb.hashnode.dev

Reading a remote log with Claude Desktop and remotify-mcp

I used Claude Desktop to read a small log on another Windows computer. Then we rejected a request to change it and checked what happened when a file was missing.

My first article covered the setup and an approved pwd command followed by a rejected date request (read the first article). For this follow-up, we prepared the receiving side again and used supervised mode throughout.

Disclosure: I work on improving remotify-mcp's visibility and have a professional connection to the project. This article reports a functional test performed on November 8, 2026. I used AI assistance during the setup and test.

Reading a file with a known result

We created a separate folder on the receiving computer and checked this eight-line synthetic test.log directly in its Bash terminal:

01 INFO Start
02 INFO Settings loaded
03 INFO Ready
04 INFO Job received
05 WARN Retry needed
06 INFO Retry successful
07 INFO Job finished
08 INFO Stop
Enter fullscreen mode Exit fullscreen mode

These lines were invented test data. The warning and retry entries did not describe actual events on either computer.

After retrieving the session details through remote_session_info, we started the supervised listener on the receiving PC. In the commands below, YOUR_USERNAME replaces the receiver's Windows username; the actual requests used the full path to our test folder.

On my own computer, I asked Claude Desktop to send exactly this command through remote_exec and show the actual output:

tail -n 5 '/c/Users/YOUR_USERNAME/remotify-naploproba-20261108/test.log'
Enter fullscreen mode Exit fullscreen mode

There were two approval steps. I clicked Allow once in Claude to permit use of the MCP tool. At the receiving computer's Run? [y/N] prompt, we entered y and pressed Enter to permit execution.

Claude returned:

04 INFO Job received
05 WARN Retry needed
06 INFO Retry successful
07 INFO Job finished
08 INFO Stop
Enter fullscreen mode Exit fullscreen mode

The response matched lines 4–8 of the file we had already checked. The listener also reported 101 bytes returned. Comparing the text with the known input gave us a direct check of this remote read.

Rejecting a write and checking the file

Next, we tested a command that would append a recognizable ninth line. The plan was to reject it on the receiver, then read the file again to check its contents.

The proposed command was:

printf '%s\n' '09 INFO Denied write test' >> '/c/Users/YOUR_USERNAME/remotify-naploproba-20261108/test.log'
Enter fullscreen mode Exit fullscreen mode

We allowed tool use in Claude, then answered n at the receiver's execution prompt. Claude eventually displayed:

[declined by operator]
Enter fullscreen mode Exit fullscreen mode

I clicked Allow once in Claude Desktop about five times before any activity was visible on the receiving computer. The chat transcript does not show whether those prompts came from one tool call or several, so I could not determine why they appeared.

We then approved a separate read through remote_exec:

cat '/c/Users/YOUR_USERNAME/remotify-naploproba-20261108/test.log'
Enter fullscreen mode Exit fullscreen mode

That returned the original eight lines. The proposed 09 INFO Denied write test entry was absent, and the existing content matched the initial file. In this test, the rejection response and the subsequent file check agreed.

Returning a missing-file error

The final check targeted missing-test.log, which we had deliberately not created in the test folder:

tail -n 5 '/c/Users/YOUR_USERNAME/remotify-naploproba-20261108/missing-test.log'
Enter fullscreen mode Exit fullscreen mode

We approved the read. Claude displayed the following error, with the username anonymized here:

tail: cannot open '/c/Users/YOUR_USERNAME/remotify-naploproba-20261108/missing-test.log' for reading: No such file or directory

[remotify: exit status 1]
Enter fullscreen mode Exit fullscreen mode

For this check, the conversation showed one tool call followed by the error. Claude did not retry or attempt to create the file. The failed read and the operator rejection therefore produced distinct responses in the conversation.

These checks covered one Windows-to-Windows setup, supervised mode and synthetic data. They show the outcomes of those specific requests; they do not establish general reliability or production security. The project documentation explains that commands run with the permissions of the account that starts the listener and that the relay stores session commands and results. Our separate test folder did not restrict those permissions.

Receiving environment: Windows, Bash 5.3.15(2), curl 8.23.0. Curl displayed a debug-build warning during the version check. Session keys are omitted, and file paths in this article use a username placeholder.

If you use MCP tools for remote work, how does your setup distinguish an operator rejection from a command failure? I would also be interested in experiences with repeated approval requests and how you investigated them.

Top comments (0)