feat(worker-mcp): raise wait_for_task_change timeout, expose queue slot state

MaxTimeoutSeconds was 170s against runs that take tens of minutes, forcing
a dozen full-context wait rounds per long-running batch. Raise it to 900s
and raise MCP_TOOL_TIMEOUT in lockstep (ClaudeProcess + every
InteractiveLaunchSpecService launch spec) to 930000ms so the client
connection actually stays open that long instead of aborting first.

Add get_queue_state (QueueStateMcpTools): configured vs. effective
parallel-slot count (via QueueService.GetSlotCountsAsync, extracted from
the former GetEffectiveMaxParallelAsync), active slots with taskId +
startedAt including the run_task_now override slot, and queued tasks in
pick order -- so a caller can observe queue occupancy instead of inferring
it from maxParallelExecutions.

src/ClaudeDo.Worker/External/QueueStateMcpTools.cs

@@ -0,0 +1,61 @@
+using System.ComponentModel;
+using ClaudeDo.Data;
+using ClaudeDo.Worker.Queue;
+using Microsoft.EntityFrameworkCore;
+using ModelContextProtocol.Server;
+using TaskStatus = ClaudeDo.Data.Models.TaskStatus;
+
+namespace ClaudeDo.Worker.External;
+
+public sealed record QueueSlotDto(string Slot, string TaskId, DateTime StartedAt);
+
+public sealed record GetQueueStateResult(
+    int ConfiguredSlots,
+    int EffectiveSlots,
+    IReadOnlyList<QueueSlotDto> ActiveSlots,
+    IReadOnlyList<string> WaitingTaskIds);
+
+[McpServerToolType]
+public sealed class QueueStateMcpTools
+{
+    private readonly QueueService _queue;
+    private readonly IDbContextFactory<ClaudeDoDbContext> _dbFactory;
+
+    public QueueStateMcpTools(QueueService queue, IDbContextFactory<ClaudeDoDbContext> dbFactory)
+    {
+        _queue = queue;
+        _dbFactory = dbFactory;
+    }
+
+    [McpServerTool, Description(
+        "Read-only snapshot of the execution queue -- observe slot occupancy instead of inferring " +
+        "it from maxParallelExecutions. Result: { configuredSlots, effectiveSlots, activeSlots: " +
+        "[{ slot, taskId, startedAt }], waitingTaskIds }. configuredSlots is Settings -> " +
+        "MaxParallelExecutions; effectiveSlots is that value stepped down by the usage throttle " +
+        "(lower when the 5h/7d usage window is filling up) -- compare the two to see whether " +
+        "throttling is currently active. activeSlots lists every task presently holding an " +
+        "execution slot, with slot \"queue\" for a normal queue slot or \"override\" for the single " +
+        "run_task_now/continue_task slot. waitingTaskIds lists queued, unblocked, non-manual, due " +
+        "tasks in the order the queue would pick them next.")]
+    public async Task<GetQueueStateResult> GetQueueState(CancellationToken cancellationToken = default)
+    {
+        var (configured, effective) = await _queue.GetSlotCountsAsync(cancellationToken);
+
+        var activeSlots = _queue.GetActive()
+            .Select(a => new QueueSlotDto(a.slot, a.taskId, a.startedAt))
+            .ToList();
+
+        await using var ctx = await _dbFactory.CreateDbContextAsync(cancellationToken);
+        var now = DateTime.UtcNow;
+        var waitingTaskIds = await ctx.Tasks
+            .Where(t => t.Status == TaskStatus.Queued
+                && t.BlockedByTaskId == null
+                && !t.IsManual
+                && (t.ScheduledFor == null || t.ScheduledFor <= now))
+            .OrderBy(t => t.SortOrder).ThenBy(t => t.CreatedAt)
+            .Select(t => t.Id)
+            .ToListAsync(cancellationToken);
+
+        return new GetQueueStateResult(configured, effective, activeSlots, waitingTaskIds);
+    }
+}

src/ClaudeDo.Worker/External/TaskWaitMcpTools.cs

@@ -12,10 +12,13 @@ public sealed record WaitForTaskChangeResult(IReadOnlyList<TaskStatusChangeDto>
 [McpServerToolType]
 public sealed class TaskWaitMcpTools
 {
-    // InteractiveLaunchSpecService sets MCP_TOOL_TIMEOUT=200000ms for the list handler
-    // session; this cap leaves a ~30s margin so the tool itself reports TimedOut instead
-    // of racing the client's own abort.
-    internal const int MaxTimeoutSeconds = 170;
+    // Every ClaudeDo-owned launcher (ClaudeProcess for headless runs, InteractiveLaunchSpecService
+    // for ConPTY sessions) sets MCP_TOOL_TIMEOUT=930000ms on the claude CLI process; this cap
+    // leaves a ~30s margin under that so the tool itself reports TimedOut instead of racing the
+    // client's own abort. A caller running claude with a different MCP_TOOL_TIMEOUT (or none --
+    // the CLI default is 60s) will see its own client-side timeout fire first; this tool has no
+    // way to detect or compensate for that from the server side.
+    internal const int MaxTimeoutSeconds = 900;
     private static readonly TimeSpan PollInterval = TimeSpan.FromMilliseconds(500);
 
     private readonly IDbContextFactory<ClaudeDoDbContext> _dbFactory;
@@ -27,9 +30,11 @@ public sealed class TaskWaitMcpTools
 
     [McpServerTool, Description(
         "Blocks until at least one of the given tasks leaves Queued/Running, or until timeoutSeconds elapses " +
-        "(clamped server-side to 170s). Returns immediately if any task is already outside Queued/Running " +
+        "(clamped server-side to 900s). Returns immediately if any task is already outside Queued/Running " +
         "when called (an unknown id is reported as status \"NotFound\" and counts as changed). Use this instead " +
-        "of polling get_task in a loop. Result: { changed: [{ taskId, status }], timedOut }.")]
+        "of polling get_task in a loop. Requires the calling claude process to run with " +
+        "MCP_TOOL_TIMEOUT >= 930000 (ms) for a long wait to actually be held open -- ClaudeDo's own " +
+        "launchers already set this. Result: { changed: [{ taskId, status }], timedOut }.")]
     public async Task<WaitForTaskChangeResult> WaitForTaskChange(
         string[] taskIds, int timeoutSeconds = 60, CancellationToken cancellationToken = default)
     {

src/ClaudeDo.Worker/Program.cs

@@ -302,6 +302,7 @@ if (cfg.ExternalMcpPort > 0)
     externalBuilder.Services.AddScoped<LifecycleMcpTools>();
     externalBuilder.Services.AddScoped<AppSettingsMcpTools>();
     externalBuilder.Services.AddScoped<TaskWaitMcpTools>();
+    externalBuilder.Services.AddScoped<QueueStateMcpTools>();
     externalBuilder.Services.AddSingleton(app.Services.GetRequiredService<AttachmentStore>());
     externalBuilder.Services.AddScoped<TaskAttachmentRepository>();
     externalBuilder.Services.AddScoped<AttachmentMcpTools>();
@@ -317,6 +318,7 @@ if (cfg.ExternalMcpPort > 0)
         .WithTools<LifecycleMcpTools>()
         .WithTools<AppSettingsMcpTools>()
         .WithTools<TaskWaitMcpTools>()
+        .WithTools<QueueStateMcpTools>()
         .WithTools<AttachmentMcpTools>();
     externalBuilder.WebHost.UseUrls($"http://127.0.0.1:{cfg.ExternalMcpPort}");
 

src/ClaudeDo.Worker/Queue/QueueService.cs

@@ -125,7 +125,7 @@ public sealed class QueueService : BackgroundService
 
                 await Task.WhenAny(wakeTask, timerTask);
 
-                var maxParallel = await GetEffectiveMaxParallelAsync(stoppingToken);
+                var (_, maxParallel) = await GetSlotCountsAsync(stoppingToken);
 
                 var gateDecision = await _usageGate.EvaluateAsync(stoppingToken);
                 await ReportUsageGateTransitionAsync(gateDecision);
@@ -200,11 +200,13 @@ public sealed class QueueService : BackgroundService
     }
 
     /// <summary>
-    /// Configured parallelism, stepped down by <see cref="UsageThrottle"/> ahead of the hard usage
-    /// gate. A missing snapshot (poll hasn't landed / endpoint unreachable) fails open to the
-    /// configured value — a broken usage poll must never stall the queue.
+    /// Configured parallelism, and that same value stepped down by <see cref="UsageThrottle"/>
+    /// ahead of the hard usage gate. A missing snapshot (poll hasn't landed / endpoint
+    /// unreachable) fails open to the configured value — a broken usage poll must never stall
+    /// the queue. Also called by <c>get_queue_state</c> (External MCP) to surface the throttle
+    /// from outside the process.
     /// </summary>
-    private async Task<int> GetEffectiveMaxParallelAsync(CancellationToken ct)
+    public async Task<(int Configured, int Effective)> GetSlotCountsAsync(CancellationToken ct)
     {
         int configured;
         int softPct, hardPct, gateFivePct, gateSevenPct;
@@ -221,14 +223,14 @@ public sealed class QueueService : BackgroundService
         catch (Exception ex)
         {
             _logger.LogWarning(ex, "Failed to read max parallel executions; defaulting to 1");
-            return 1;
+            return (1, 1);
         }
 
         var snapshot = _usageState.Snapshot;
         if (snapshot is null || _usageState.LastError is not null)
         {
             _lastEffectiveSlots = configured;
-            return configured;
+            return (configured, configured);
         }
 
         var effective = UsageThrottle.EffectiveSlots(
@@ -236,7 +238,7 @@ public sealed class QueueService : BackgroundService
             softPct, hardPct, gateFivePct, gateSevenPct);
 
         ReportThrottleTransition(configured, effective, snapshot);
-        return effective;
+        return (configured, effective);
     }
 
     private void ReportThrottleTransition(int configured, int effective, UsageSnapshot snapshot)

src/ClaudeDo.Worker/Runner/ClaudeProcess.cs

@@ -40,9 +40,11 @@ public sealed class ClaudeProcess : IClaudeProcess
             psi.ArgumentList.Add(arg);
 
         // Claude Code caps HTTP MCP tool calls at 60 s unless MCP_TOOL_TIMEOUT is raised.
-        // The in-task AskUser tool blocks up to 3 min waiting for the user, so lift the cap
-        // (with margin) or that wait would be killed early. Harmless for every other tool.
-        psi.Environment["MCP_TOOL_TIMEOUT"] = "200000";
+        // wait_for_task_change blocks up to TaskWaitMcpTools.MaxTimeoutSeconds (900 s) and the
+        // in-task AskUser tool blocks up to 3 min waiting for the user, so lift the cap well
+        // past the longer of the two (with margin) or that wait would be killed early. Harmless
+        // for every other tool. Keep in sync with InteractiveLaunchSpecService's MCP_TOOL_TIMEOUT.
+        psi.Environment["MCP_TOOL_TIMEOUT"] = "930000";
 
         using var process = new Process { StartInfo = psi };
         process.Start();

src/ClaudeDo.Worker/Runner/InteractiveLaunchSpecService.cs

@@ -14,7 +14,7 @@ namespace ClaudeDo.Worker.Runner;
 // session in an existing task's worktree -- the SAME worktree prep as an autonomous run:
 // session-skills seeded onto disk (reuses ISessionSkillSeeder + TaskRunner.UnionSkillNames,
 // exactly like TaskRunner.RunAsync/ContinueAsync) and the same run environment variables
-// (reuses ClaudeProcess's MCP_TOOL_TIMEOUT). Exe/Args reuse WindowsTerminalLauncher's
+// (matches ClaudeProcess's MCP_TOOL_TIMEOUT). Exe/Args reuse WindowsTerminalLauncher's
 // --resume argument construction. Guards: no running/queued task, and (once a worktree
 // exists) it must be live on disk -- but a never-run task (no persisted SessionId) is not
 // an error here, it's a fresh-start spec.
@@ -107,11 +107,11 @@ public sealed class InteractiveLaunchSpecService : IInteractiveLaunchSpecService
             : await BuildFreshTaskArgsAsync(task, effort, ct);
 
         // Same run environment variable ClaudeProcess sets for every headless run: the
-        // AskUser MCP tool call caps at 60s unless raised, and lifting it is harmless for
-        // every other tool.
+        // AskUser MCP tool call and wait_for_task_change cap at 60s unless raised, and lifting
+        // it is harmless for every other tool. Keep in sync with ClaudeProcess's MCP_TOOL_TIMEOUT.
         var env = new Dictionary<string, string>
         {
-            ["MCP_TOOL_TIMEOUT"] = "200000",
+            ["MCP_TOOL_TIMEOUT"] = "930000",
         };
 
         return new LaunchSpec(worktree.Path, resolvedClaude, args, env);
@@ -129,7 +129,7 @@ public sealed class InteractiveLaunchSpecService : IInteractiveLaunchSpecService
         {
             ["MAX_THINKING_TOKENS"] = "20000",
             ["CLAUDEDO_PLANNING_TOKEN"] = ctx.Token,
-            ["MCP_TOOL_TIMEOUT"] = "200000",
+            ["MCP_TOOL_TIMEOUT"] = "930000",
         };
 
         return new LaunchSpec(
@@ -147,7 +147,7 @@ public sealed class InteractiveLaunchSpecService : IInteractiveLaunchSpecService
         var env = new Dictionary<string, string>
         {
             ["CLAUDEDO_PLANNING_TOKEN"] = ctx.Token,
-            ["MCP_TOOL_TIMEOUT"] = "200000",
+            ["MCP_TOOL_TIMEOUT"] = "930000",
         };
 
         return new LaunchSpec(
@@ -170,7 +170,7 @@ public sealed class InteractiveLaunchSpecService : IInteractiveLaunchSpecService
 
         var env = new Dictionary<string, string>
         {
-            ["MCP_TOOL_TIMEOUT"] = "200000",
+            ["MCP_TOOL_TIMEOUT"] = "930000",
         };
 
         // No task and no list here — the global default model's preset decides the effort.
@@ -247,7 +247,7 @@ public sealed class InteractiveLaunchSpecService : IInteractiveLaunchSpecService
 
         var env = new Dictionary<string, string>
         {
-            ["MCP_TOOL_TIMEOUT"] = "200000",
+            ["MCP_TOOL_TIMEOUT"] = "930000",
         };
 
         return new LaunchSpec(repoDir, resolvedClaude, args, env);