06 Jun '25
Agentic Elixir superpowers: Zed + Tidewave + AshAI
For a few years now whenever I need to build any sort of networked interactive experience, I reach for Elixir (with Phoenix LiveView and Ash). It’s an extremely productive combination, especially when you want to do more complex client/server information flows than the standard request/response UX. I’ve also (for the last year or two) ditched Emacs for Zed. Again, if you take the time to master these tools I think they’re excellent.
Like every indie hacker and their wombat I’m experimenting with LLM Agents as part of my software development workflow. The new frontier (which I’m excited about, but ask me in a few months how it’s gone) is putting all of the above together. If you’d like to do that too, and want to see a real-world example of how these things can be plugged into one another, here’s my setup.
Then, my Zed settings file has this MCP server configuration:
"context_servers": {
"ash_ai": {
"command": {
"path": "mcp-proxy",
"args": ["http://localhost:4000/ash_ai/mcp"],
"env": {}
},
"settings": {}
},
"tidewave": {
"command": {
"path": "mcp-proxy",
"args": ["http://localhost:4000/tidewave/mcp"],
"env": {}
},
"settings": {}
}
},
Even after following the instructions for those tools, you’ll need a way to set up all the proxies and pipe everything together. I wrote this script, which you’re free to use (MIT Licence) if it helps.
Put it on your ~PATH and then (in your Phoenix project root) run it like so:
[16:03] daysy:panic $ tidewave-proxy.sh
🚀 MCP Proxy Development Environment
====================================
Configuration:
Host: localhost
Port: 4000
Ash AI MCP: http://localhost:4000/ash_ai/mcp
Tidewave MCP: http://localhost:4000/tidewave/mcp
🔗 Starting mcp-proxy instances...
Starting mcp-proxy for ash_ai...
✅ mcp-proxy for ash_ai started (PID: 88815)
Log file: /tmp/mcp_ash_ai.log
Starting mcp-proxy for tidewave...
✅ mcp-proxy for tidewave started (PID: 88816)
Log file: /tmp/mcp_tidewave.log
🎉 MCP Proxy environment is ready!
=====================================
Available MCP Servers:
📊 Ash AI MCP: http://localhost:4000/ash_ai/mcp
🌊 Tidewave MCP: http://localhost:4000/tidewave/mcp
MCP Proxy Commands:
For Ash AI: mcp-proxy http://localhost:4000/ash_ai/mcp
For Tidewave: mcp-proxy http://localhost:4000/tidewave/mcp
Next Steps:
1. Configure your MCP client (Zed, Claude Desktop, etc.)
2. Use the endpoints above with your MCP client
3. Test with: curl -H 'Accept: application/json' <endpoint>
Log Files:
Ash AI Proxy: /tmp/mcp_ash_ai.log
Tidewave Proxy: /tmp/mcp_tidewave.log
Press Ctrl+C to stop all services and clean up
🔥 Starting Phoenix development server with REPL...
==================================================
Erlang/OTP 27 [erts-15.2.5] [source] [64-bit] [smp:16:16] [ds:16:16:10] [async-threads:1] [jit]
[info] Migrations already up
[info] Running PanicWeb.Endpoint with Bandit 1.7.0 at 127.0.0.1:4000 (http)
[info] Access PanicWeb.Endpoint at http://localhost:4000
Interactive Elixir (1.18.4) - press Ctrl+C to exit (type h() ENTER for help)
[watch] build finished, watching for changes...
Rebuilding...
Done in 196ms.
Then, to test, in the Agent panel try something like:
Use the tidewave project eval tool to add 10+15 in Elixir.
If everything’s hooked up right, you’ll get the answer.
One other tip: make sure you’re using OTP27 rather than the (newest) OTP28—there’s an error on one of the AshAI deps which stops it compiling on the latest OTP. I suspect it’ll be fixed soon, though.