MCP Server for Tauri

[velvet mango]

Let your AI agent access your Tauri app to take screenshots, tap/click/swipe/scroll, read device, console logs, and IPC calls for debugging/tracing, and more.

A few ways you can use it (example prompts):

Improved UI Development
> Add a Login button to the navigation bar. Check your work with screenshots and element inspection with the Tauri MCP. Don't stop until it looks correct.

QA Testing
> Use the Tauri MCP to give my app a full QA test. Resize the window to test multiple layout sizes.

Debugging
> What's causing the error in the webview console?

More here: https://hypothesi.github.io/mcp-server-tauri/
GitHub repo: https://github.com/hypothesi/mcp-server-tauri

[zinc kraken]

Super useful, I can now have AI agent debug itself with much less intervention or work from me in a separate browser. Was able to retool some debug services I had built in the app from OpenAPI to MCP instead to unify the debuggin path.

[velvet mango]

I’m glad you found it helpful! I use it that way too. I found that having a rule that tells the agent to use it to check its work before finishing dramatically improves the quality of the final output.

If you have any ideas for improvements or new tools or bug reports, let me know!

[civic fossil]

This is awesome, just plugged it in to Claude Code. Leaving a star. 🫶

[zinc kraken]

As with all things agentic, rules into AGENTS.md or a SKILL.md definitely help to certify the process and outcome. For debugging I'm already seeing Claude proactively use this MCP to check its work or investigate something I have said is a problem using a guided workflow and preferred MCP actions I have laid out.

[versed oasis]

There were a couple of other tauri-mcp projects out there that didn't have a lot of polish and one I remember that just flat out didnt work. Looking forward to testing this one out!

[velvet mango]

Great! If you find something that doesn’t work, just let me know and we can get it fixed. I’m around.

[exotic willow]

Didn't try it yet but seems cool, gave you a star on GitHub (⁠ ⁠◜⁠‿⁠◝⁠ ⁠)⁠♡

[nocturne rune]

Going to test this out today with Claude code to see if it can pull off UI testing

[velvet mango]

Let me know if you run into any snags or if you have feature requests!

One improvement I’d like to make soon is to add a tool that returns a concise representation of the DOM, similar to Chrome DevTools MCP’s Accessibility Tree tool.

The agent sometimes has trouble querying DOM elements because it doesn’t first know what selectors already exist, and often does not realize it can use existing tools to traverse the DOM and “look around”.

That may also cut down on how often it relies on screenshots to check its work.

[versed oasis]

Setup was pretty seamless on windows vscode! Gunna try out some of the functionality now

[versed oasis]

any chance you'd mind sharing some examples of your rules (as related to tauri mcp)?

[chrome tree]

it works with Gemini CLI too 👍

[zinc kraken]

Just create a debug skill for your agent, and in that give concrete examples of the MCP commands to use to exercise your app.

Updated screenshot ability works well, screenshots have opened autonomy for agent to work through items without manual intervention.

[velvet mango]

I also shipped the improvement I mentioned above that gives the agent some tools to get an overview of the DOM and Accessibility tree structure.

It can save tokens vs a screenshot when the agent is able to use it to check its work.

It’s a good thing to reference in the skill in addition to the screenshot tool. Sometimes agents have trouble recognizing finer details like px alignment issues in screenshots but can work it out if they realize they can query the DOM and the elements’ computed styles while debugging.

[zinc kraken]

yeah its working nicely now, can watch the agent spin up Tauri MCP and work through its questions about state in debugging. My app is data intensive so I've got lots of back and forth from Rust backend to React front end to decipher, a simple browser only solution wasn't cutting it at all.

[odd whale]

It works smoothly.
I've supplied a PR because in some case the bridge was preventing to get some relevant warning or errors. https://github.com/hypothesi/mcp-server-tauri/issues/9