diff --git a/src/main_ghostty.zig b/src/main_ghostty.zig
index 6354214d0..6089371c2 100644
--- a/src/main_ghostty.zig
+++ b/src/main_ghostty.zig
@@ -18,6 +18,7 @@ const fontconfig = @import("fontconfig");
 const harfbuzz = @import("harfbuzz");
 const renderer = @import("renderer.zig");
 const apprt = @import("apprt.zig");
+const tcp = @import("tcp.zig");
 
 const App = @import("App.zig");
 const Ghostty = @import("main_c.zig").Ghostty;
@@ -112,6 +113,18 @@ pub fn main() !MainReturn {
     // when the first surface is created.
     if (@hasDecl(apprt.App, "startQuitTimer")) app_runtime.startQuitTimer();
 
+    // Not sure where this should go tbh
+    var tcp_thread = try tcp.Thread.init(alloc);
+    defer tcp_thread.deinit();
+
+    var tcp_thr = try std.Thread.spawn(
+        .{},
+        tcp.Thread.threadMain,
+        .{&tcp_thread},
+    );
+
+    tcp_thr.setName("tcp") catch {};
+
     // Run the GUI event loop
     try app_runtime.run();
 }
diff --git a/src/tcp.zig b/src/tcp.zig
new file mode 100644
index 000000000..68450920b
--- /dev/null
+++ b/src/tcp.zig
@@ -0,0 +1,9 @@
+//! TCP implementation. The TCP implementation is responsible for
+//! responding to TCP requests and dispatching them to the app's Mailbox.
+pub const Thread = @import("tcp/Thread.zig");
+pub const Server = @import("tcp/Server.zig");
+pub const Command = @import("tcp/Command.zig").Command;
+
+test {
+    @import("std").testing.refAllDecls(@This());
+}
diff --git a/src/tcp/Command.zig b/src/tcp/Command.zig
new file mode 100644
index 000000000..65a9a92de
--- /dev/null
+++ b/src/tcp/Command.zig
@@ -0,0 +1,50 @@
+const std = @import("std");
+const log = std.log.scoped(.tcp_thread);
+
+pub const Command = enum {
+    ping,
+
+    pub const Error = error{
+        InvalidInput,
+        InvalidCommand,
+    };
+
+    /// Takes the bytes from our TCP handle and parses them into a Command.
+    pub fn parse(raw_read: []const u8) !Command {
+        const trimmed = std.mem.trim(u8, raw_read, "\t\n\r");
+
+        if (trimmed.len == 0) {
+            return error.InvalidInput;
+        }
+
+        var iter = std.mem.splitScalar(u8, trimmed, ' ');
+
+        // Not using .first() because it returns the remainder of the slice
+        // Instead we're doing what's equivalent to popping the first element
+        const cmdName = iter.next() orelse return error.InvalidInput;
+        log.debug("got command name={s}", .{cmdName});
+        // TODO: Handle/support arguments
+
+        return std.meta.stringToEnum(Command, cmdName) orelse return error.InvalidCommand;
+    }
+
+    pub fn handle(self: Command) ![]const u8 {
+        switch (self) {
+            .ping => return ping(),
+        }
+    }
+
+    pub fn handleError(err: Command.Error) ![]const u8 {
+        switch (err) {
+            error.InvalidInput => return "INVALID INPUT\n",
+            error.InvalidCommand => return "INVALID COMMAND\n",
+        }
+    }
+};
+
+// Want to move this to different file, not sure how I want it organized yet
+fn ping() []const u8 {
+    return "PONG\n";
+}
+
+// TODO: These need proper testing.
diff --git a/src/tcp/Server.zig b/src/tcp/Server.zig
new file mode 100644
index 000000000..0568ed11d
--- /dev/null
+++ b/src/tcp/Server.zig
@@ -0,0 +1,317 @@
+const std = @import("std");
+const xev = @import("xev");
+const tcp = @import("../tcp.zig");
+
+const reject_client = @import("./handlers/reject.zig").reject_client;
+
+const Allocator = std.mem.Allocator;
+const CompletionPool = std.heap.MemoryPool(xev.Completion);
+const SocketPool = std.heap.MemoryPool(xev.TCP);
+const BufferPool = std.heap.MemoryPool([1024]u8);
+const log = std.log.scoped(.tcp_thread);
+
+/// Maximum connections we allow at once
+const MAX_CLIENTS = 2;
+
+/// Acceptor polling rate in milliseconds
+const ACCEPTOR_RATE = 1;
+
+/// A wrapper around the TCP server socket
+pub const Server = @This();
+
+/// Memory pool that stores the completions required by xev
+comp_pool: CompletionPool,
+
+/// Memory pool that stores client sockets
+sock_pool: SocketPool,
+
+/// Memory pool that stores buffers for reading and writing
+buf_pool: BufferPool,
+
+/// Stop flag
+stop: bool,
+
+/// Event loop
+loop: xev.Loop,
+
+/// Server socket
+socket: xev.TCP,
+
+/// Timer for accepting connections
+acceptor: xev.Timer,
+
+/// Array of client sockets
+clients: [MAX_CLIENTS]*xev.TCP,
+
+/// Number of clients connected
+clients_count: usize,
+
+/// Initializes the server with the given allocator and address
+pub fn init(alloc: Allocator, addr: std.net.Address) !Server {
+    const socket = try xev.TCP.init(addr);
+    try socket.bind(addr);
+
+    return Server{
+        .comp_pool = CompletionPool.init(alloc),
+        .sock_pool = SocketPool.init(alloc),
+        .buf_pool = BufferPool.init(alloc),
+        .stop = false,
+        .loop = try xev.Loop.init(.{}),
+        .socket = socket,
+        .acceptor = try xev.Timer.init(),
+        .clients = undefined,
+        .clients_count = 0,
+    };
+}
+
+/// Deinitializes the server
+pub fn deinit(self: *Server) void {
+    self.buf_pool.deinit();
+    self.comp_pool.deinit();
+    self.sock_pool.deinit();
+    self.loop.deinit();
+    self.acceptor.deinit();
+}
+
+/// Starts the timer which tries to accept connections
+pub fn start(self: *Server) !void {
+    try self.socket.listen(MAX_CLIENTS);
+    log.info("bound server to socket={any}", .{self.socket});
+
+    // Each acceptor borrows a completion from the pool
+    // We do this because the completion is passed to the client TCP handlers
+    const c = self.comp_pool.create() catch {
+        log.err("couldn't allocate completion in pool", .{});
+        return error.OutOfMemory;
+    };
+
+    self.acceptor.run(&self.loop, c, ACCEPTOR_RATE, Server, self, acceptor);
+    while (!self.stop) {
+        try self.loop.run(.until_done);
+    }
+}
+
+/// Convenience function to destroy a buffer in our pool
+fn destroyBuffer(self: *Server, buf: []const u8) void {
+    self.buf_pool.destroy(@alignCast(
+        @as(*[1024]u8, @ptrFromInt(@intFromPtr(buf.ptr))),
+    ));
+}
+
+/// This runs on a loop and attempts to accept a connection to pass into the
+/// handlers that manage client communication. It is very important to note
+/// that the xev.Completion is shared between the acceptor and other handlers.
+/// When a client disconnects, only then the completion is freed.
+fn acceptor(
+    self_: ?*Server,
+    loop: *xev.Loop,
+    c: *xev.Completion,
+    e: xev.Timer.RunError!void,
+) xev.CallbackAction {
+    const self = self_.?;
+    e catch {
+        log.err("timer error", .{});
+        return .disarm;
+    };
+
+    self.socket.accept(loop, c, Server, self, acceptHandler);
+
+    // We need to create a new completion for the next acceptor since each
+    // TCP connection will need its own if it successfully accepts.
+    const accept_recomp = self.comp_pool.create() catch {
+        log.err("couldn't allocate completion in pool", .{});
+        return .disarm;
+    };
+
+    // We can't rearm because it'll repeat *this* instance of the acceptor
+    // So if the socket fails to accept it won't actually accept anything new
+    self.acceptor.run(loop, accept_recomp, ACCEPTOR_RATE, Server, self, acceptor);
+    return .disarm;
+}
+
+/// Accepts a new client connection and starts reading from it until EOF.
+fn acceptHandler(
+    self_: ?*Server,
+    l: *xev.Loop,
+    c: *xev.Completion,
+    e: xev.TCP.AcceptError!xev.TCP,
+) xev.CallbackAction {
+    const self = self_.?;
+    const sock = self.sock_pool.create() catch {
+        log.err("couldn't allocate socket in pool", .{});
+        return .disarm;
+    };
+
+    sock.* = e catch {
+        log.err("accept error", .{});
+        self.sock_pool.destroy(sock);
+        return .disarm;
+    };
+
+    const comp_w = self.comp_pool.create() catch {
+        log.err("couldn't allocate completion in pool", .{});
+        return .disarm;
+    };
+
+    const buf_w = self.buf_pool.create() catch {
+        log.err("couldn't allocate buffer in pool", .{});
+        self.sock_pool.destroy(sock);
+        return .disarm;
+    };
+
+    if (self.clients_count == MAX_CLIENTS) {
+        log.warn("max clients reached, rejecting fd={d}", .{sock.fd});
+        reject_client(self, sock) catch return .rearm;
+        return .disarm;
+    }
+
+    log.info("accepted connection fd={d}", .{sock.fd});
+    self.clients[self.clients_count] = sock;
+    self.clients_count += 1;
+
+    const buf_r = self.buf_pool.create() catch {
+        log.err("couldn't allocate buffer in pool", .{});
+        return .disarm;
+    };
+
+    @memcpy(buf_w.ptr, "GHOSTTY\n");
+    sock.write(l, comp_w, .{ .slice = buf_w }, Server, self, writeHandler);
+    sock.read(l, c, .{ .slice = buf_r }, Server, self, readHandler);
+    return .disarm;
+}
+
+fn readHandler(
+    self_: ?*Server,
+    loop: *xev.Loop,
+    comp: *xev.Completion,
+    sock: xev.TCP,
+    buf: xev.ReadBuffer,
+    e: xev.TCP.ReadError!usize,
+) xev.CallbackAction {
+    const self = self_.?;
+    const l = e catch |err| {
+        switch (err) {
+            error.EOF => {
+                log.info("disconnecting client fd={d}", .{sock.fd});
+                destroyBuffer(self, buf.slice);
+                self.clients_count -= 1;
+                sock.close(loop, comp, Server, self, closeHandler);
+                return .disarm;
+            },
+
+            else => {
+                destroyBuffer(self, buf.slice);
+                log.err("read error", .{});
+                return .disarm;
+            },
+        }
+    };
+
+    const comp_w = self.comp_pool.create() catch {
+        log.err("couldn't allocate completion in pool", .{});
+        return .rearm;
+    };
+
+    const buf_w = self.buf_pool.create() catch {
+        log.err("couldn't allocate buffer in pool", .{});
+        return .rearm;
+    };
+
+    // TODO: What should we do with the rest of the buffer?
+    var iter = std.mem.splitScalar(u8, buf.slice[0..l], '\n');
+    while (iter.next()) |item| {
+        if (item.len == 0) {
+            continue;
+        }
+
+        const c = tcp.Command.parse(item) catch |err| {
+            const res = try tcp.Command.handleError(err);
+            @memcpy(buf_w.ptr, res.ptr[0..res.len]);
+            sock.write(loop, comp_w, .{ .slice = buf_w }, Server, self, writeHandler);
+            return .rearm;
+        };
+
+        const res = try tcp.Command.handle(c);
+        @memcpy(buf_w.ptr, res.ptr[0..res.len]);
+    }
+
+    sock.write(loop, comp_w, .{ .slice = buf_w }, Server, self, writeHandler);
+    return .rearm;
+}
+
+fn writeHandler(
+    self_: ?*Server,
+    loop: *xev.Loop,
+    comp: *xev.Completion,
+    sock: xev.TCP,
+    buf: xev.WriteBuffer,
+    e: xev.TCP.WriteError!usize,
+) xev.CallbackAction {
+    _ = sock;
+    _ = comp;
+    _ = loop;
+    const self = self_.?;
+    _ = e catch |err| {
+        destroyBuffer(self, buf.slice);
+        log.err("write error {any}", .{err});
+        return .disarm;
+    };
+
+    return .disarm;
+}
+
+fn rejectHandler(
+    self_: ?*Server,
+    l: *xev.Loop,
+    c: *xev.Completion,
+    sock: xev.TCP,
+    buf: xev.WriteBuffer,
+    e: xev.TCP.WriteError!usize,
+) xev.CallbackAction {
+    const self = self_.?;
+    _ = e catch |err| {
+        destroyBuffer(self, buf.slice);
+        log.err("write error {any}", .{err});
+        return .disarm;
+    };
+
+    sock.close(l, c, Server, self, closeHandler);
+    return .disarm;
+}
+
+fn shutdownHandler(
+    self_: ?*Server,
+    loop: *xev.Loop,
+    comp: *xev.Completion,
+    sock: xev.TCP,
+    e: xev.TCP.ShutdownError!void,
+) xev.CallbackAction {
+    e catch {
+        // Is this even possible?
+        log.err("couldn't shutdown socket", .{});
+    };
+
+    const self = self_.?;
+
+    sock.close(loop, comp, Server, self, closeHandler);
+    return .disarm;
+}
+
+pub fn closeHandler(
+    self_: ?*Server,
+    loop: *xev.Loop,
+    comp: *xev.Completion,
+    sock: xev.TCP,
+    e: xev.TCP.CloseError!void,
+) xev.CallbackAction {
+    _ = sock;
+    _ = loop;
+
+    e catch {
+        log.err("couldn't close socket", .{});
+    };
+
+    const self = self_.?;
+    self.comp_pool.destroy(comp);
+    return .disarm;
+}
diff --git a/src/tcp/Thread.zig b/src/tcp/Thread.zig
new file mode 100644
index 000000000..3e0f86a9a
--- /dev/null
+++ b/src/tcp/Thread.zig
@@ -0,0 +1,50 @@
+//! Represents the libxev thread for handling TCP connections.
+pub const Thread = @This();
+
+const std = @import("std");
+const xev = @import("xev");
+const tcp = @import("../tcp.zig");
+const App = @import("../App.zig");
+
+const Allocator = std.mem.Allocator;
+const log = std.log.scoped(.tcp_thread);
+
+/// Allocator used for xev allocations.
+alloc: std.mem.Allocator,
+
+/// The TCP server for handling incoming connections.
+server: tcp.Server,
+
+/// Initialize the thread. This does not START the thread. This only sets
+/// up all the internal state necessary prior to starting the thread. It
+/// is up to the caller to start the thread with the threadMain entrypoint.
+pub fn init(alloc: Allocator) !Thread {
+    // TODO: Configurable addresses and socket paths
+    const addr = try std.net.Address.parseIp4("127.0.0.1", 9090);
+    var server = try tcp.Server.init(alloc, addr);
+    errdefer server.deinit();
+
+    return Thread{
+        .alloc = alloc,
+        .server = server,
+    };
+}
+
+/// Clean up the thread. This is only safe to call once the thread
+/// completes executing; the caller must join prior to this.
+pub fn deinit(self: *Thread) void {
+    self.server.deinit();
+}
+
+/// The main entrypoint for the thread.
+pub fn threadMain(self: *Thread) void {
+    self.threadMain_() catch |err| {
+        log.warn("error in tcp thread err={any}", .{err});
+    };
+}
+
+fn threadMain_(self: *Thread) !void {
+    log.debug("starting tcp thread", .{});
+    defer log.debug("tcp thread exited", .{});
+    try self.server.start();
+}
diff --git a/src/tcp/commands/ping.zig b/src/tcp/commands/ping.zig
new file mode 100644
index 000000000..e69de29bb
diff --git a/src/tcp/handlers/reject.zig b/src/tcp/handlers/reject.zig
new file mode 100644
index 000000000..85283f802
--- /dev/null
+++ b/src/tcp/handlers/reject.zig
@@ -0,0 +1,47 @@
+const xev = @import("xev");
+const std = @import("std");
+const Server = @import("../Server.zig").Server;
+
+const log = std.log.scoped(.tcp_thread);
+
+fn destroyBuffer(self: *Server, buf: []const u8) void {
+    self.buf_pool.destroy(@alignCast(
+        @as(*[1024]u8, @ptrFromInt(@intFromPtr(buf.ptr))),
+    ));
+}
+
+const RejectError = error{
+    AllocError,
+    WriteError,
+};
+
+// TODO: Support client rejection reasons.
+pub fn reject_client(self: *Server, c: *xev.TCP) !void {
+    const buf_w = self.buf_pool.create() catch return RejectError.AllocError;
+    const comp_w = self.comp_pool.create() catch {
+        destroyBuffer(self, buf_w);
+        return RejectError.AllocError;
+    };
+
+    @memcpy(buf_w.ptr, "ERR: Max connections reached\n");
+    c.write(&self.loop, comp_w, .{ .slice = buf_w }, Server, self, wHandler);
+}
+
+fn wHandler(
+    self_: ?*Server,
+    l: *xev.Loop,
+    c: *xev.Completion,
+    client: xev.TCP,
+    b: xev.WriteBuffer,
+    e: xev.TCP.WriteError!usize,
+) xev.CallbackAction {
+    const self = self_.?;
+    _ = e catch |err| {
+        destroyBuffer(self, b.slice);
+        log.err("write error {any}", .{err});
+        return .disarm;
+    };
+
+    client.close(l, c, Server, self, Server.closeHandler);
+    return .disarm;
+}