[BUG] Cookie authentication fails with "cookie may be expired or invalid" on Windows

[arturochian]

Environment Details

• **Operating System:**Windows 10 Home 22H2
• Node.js Version: v22.17.0
• Python Version: 3.13.51420396 2025-07-09)
• UV Version: 0.7.20 (251420396 2025-07-09)
• Chrome for Testing Version: 138.0.7204.94 (Build oficial) (64 bits)

Bug Description

The MCP server fails to authenticate with LinkedIn using the cookie method. Despite a correct setup, the process consistently fails at the final authentication step, returning the error Login failed: Cookie authentication failed - cookie may be expired or invalid, even when using a freshly generated li_at cookie from a clean session.

Steps to Reproduce

1. Cloned the repository on Windows.
2. Installed uv using the recommended PowerShell script (irm https://astral.sh/uv/install.ps1 | iex).
3. In the project directory, created and activated a virtual environment: uv venv and .\.venv\Scripts\Activate.ps1.
4. Installed all dependencies successfully: uv sync and uv sync --group dev.
5. Downloaded the matching "Chrome for Testing" browser and chromedriver.exe.
6. Placed both chrome.exe and chromedriver.exe in the same directory (D:\ABCN\chrome-win64) and added this directory to the system's Path.
7. Set the LINKEDIN_COOKIE environment variable with a fresh li_at cookie obtained after logging out of all other sessions.
8. Ran the server manually with debug logging: uv run main.py --no-headless --no-lazy-init --log-level DEBUG.

Expected Behavior

The server should use the provided cookie, successfully authenticate the session in the automated "Chrome for Testing" browser, and become ready to accept commands.

Actual Behavior / Logs

The server starts, correctly reads the environment variables, finds the chromedriver, and launches the "Chrome for Testing" browser successfully. The WebDriver initializes correctly. However, the final authentication step fails.

The log shows the following critical error:

drivers.chrome - WARNING - Cookie authentication failed: Cookie login failed - cookie may be expired or invalid
main - ERROR - Failed to initialize driver: Login failed: Cookie authentication failed - cookie may be expired or invalid

Additional Context & Troubleshooting

• The issue is not a port conflict (EADDRINUSE was resolved).
• The issue is not the Path. We confirmed the script finds and uses the correct chromedriver.exe.
• We also tried explicitly setting the CHROMEDRIVER_PATH environment variable, with the same result.
• The failure seems to indicate that LinkedIn's current security measures may be detecting and invalidating the session when initiated by an automated browser, even if the cookie itself is valid.

Thank you for looking into this.

[stickerdaniel]

Hey, the login most likely fails because the cookie is from a Windows session, but the scraper is using a macOS user agent. You can run the scraper with your password/email instead of cookie login, and it should log in with the scraper user agent and store the cookie from that login for the next usage.

[stickerdaniel]

Let me know if this issue is resolved in the latest version / after pulling the changes. Also, make sure that you aren't logged in with the same cookie in your browser or a different instance at the same time you're using the MCP server.

[arturochian]

Hi Daniel,

Thanks a lot for the quick and detailed response. The explanation regarding the User-Agent mismatch makes perfect sense and clarifies the issue I was seeing.

I will test the email/password login method with the latest version as you suggested, ensuring all other browser sessions are closed.

I should have some time to test this over the next couple of days, and I'll report back here with the outcome.

Thanks again for your support!

[arturochian]

Hi Daniel,

Thanks again for your detailed suggestions.

I've followed all the steps you recommended. I pulled the latest version of the repository, re-synced all dependencies with uv, and set up the matching Chrome for Testing and ChromeDriver with the correct environment variables.

Unfortunately, when trying to trigger the authentication from within Gemini CLI, it now consistently fails. The command @linkedin.authenticate_LinkedIn results in a 400 Bad Request from the Gemini API itself:

[API Error: { code: 400, message: "Request contains an invalid argument.", status: "INVALID_ARGUMENT" }]

Note: this message is new one.

It seems that even with the server working, the Gemini CLI is unable to correctly formulate the tool call to the MCP.

Let me know if there are any other diagnostics I can provide. Thanks again for your help.

[stickerdaniel]

https://github.com/google-gemini/gemini-cli/blob/main/docs/tools/mcp-server.md

[stickerdaniel]

Here is a video showing how to use this MCP with Gemini CLI!