@kolbo/mcp 1.79.2 → 1.79.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kolbo/mcp",
-  "version": "1.79.2",
+  "version": "1.79.3",
   "description": "Kolbo AI MCP Server - Generate images, videos, music, speech, and sound effects from Claude Code",
   "main": "src/index.js",
   "bin": {

package/src/progress.js

@@ -77,6 +77,19 @@ function trackGeneration(id, cancel) {
   if (extra.signal.aborted) cancelTracked();
 }
 
+/**
+ * Remove a generation from the tracked set so it will NOT be cancelled when
+ * the MCP host aborts the tool call. Use this when the CLIENT-SIDE poll
+ * window times out but the generation should keep running server-side — the
+ * tool returns early with generation_id + _timed_out:true, and the caller
+ * re-issues get_generation_status to collect the result later.
+ */
+function untrackGeneration(id) {
+  const extra = storage.getStore();
+  if (!extra?.__kolboTrackedGenerations || !id) return;
+  extra.__kolboTrackedGenerations.delete(String(id));
+}
+
 async function generation(id) {
   const extra = storage.getStore();
   const token = extra?._meta?.progressToken;
@@ -123,4 +136,4 @@ async function sendGenerationProgress(extra, token, id) {
   });
 }
 
-module.exports = { run, generation, tick, signal, wait, trackGeneration };
+module.exports = { run, generation, tick, signal, wait, trackGeneration, untrackGeneration };

package/src/tools/_shared.js

@@ -290,7 +290,13 @@ async function resolveToBuffer(source, kind, opts = {}) {
 // it instead of duplicating the try/catch. Genuine failures (state
 // failed/cancelled → GenerationFailedError) are NOT caught here; they still
 // throw and surface as a real tool error.
+//
+// CRITICAL: when the poll window times out, we UNTRACK the generation so that
+// when the MCP host eventually aborts the tool call (e.g., at ~180s), it does
+// NOT cancel the server-side generation. The generation keeps running, and the
+// caller can collect it later with get_generation_status.
 const { pollUntilDone } = require('../polling');
+const progress = require('../progress');
 
 /**
  * @returns {Promise<{result: object}|{timedOut: object}>}
@@ -301,6 +307,8 @@ async function pollOrTimedOut(client, generationId, pollOpts) {
     return { result: await pollUntilDone(client, generationId, pollOpts) };
   } catch (err) {
     if (!err || !err.timedOut) throw err;
+    // Untrack so the MCP host aborting this tool call does NOT cancel the generation.
+    progress.untrackGeneration(generationId);
     return {
       timedOut: {
         content: [{

package/src/tools/chat.js

@@ -43,9 +43,9 @@ function registerChatTools(server, client) {
         project_id
       });
 
-      // Deep think reasoning can run far longer than normal chat. Also grant
-      // extra time when web_search is on (may fetch + analyze multiple pages).
-      const timeout = deep_think ? 600000 : (web_search ? 240000 : 120000);
+      // Return before host cutoff (150s) even for deep think / web search.
+      // The LLM can call this tool again with the same session_id to continue.
+      const timeout = 150000;
 
       const poll = await pollOrTimedOut(client, gen.message_id, {
         interval: (gen.poll_interval_hint || 2) * 1000,

package/src/tools/generate.js

@@ -245,7 +245,7 @@ function registerGenerateTools(server, client, options = {}) {
           status_args: { generation_ids: batch.ids, wait: true },
           reference_images
         });
-        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 3) * 1000, timeout: 240000 }, 'generate_image');
+        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 3) * 1000, timeout: 150000 }, 'generate_image');
       }
 
       const gen = await client.post('/v1/generate/image', { ...shared, prompt, num_images });
@@ -258,7 +258,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 3) * 1000,
-        timeout: 120000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -328,7 +328,7 @@ function registerGenerateTools(server, client, options = {}) {
           status_args: { generation_ids: batch.ids, wait: true },
           reference_images: source_images
         });
-        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 3) * 1000, timeout: 240000 }, 'generate_image_edit');
+        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 3) * 1000, timeout: 150000 }, 'generate_image_edit');
       }
 
       const gen = await client.post('/v1/generate/image-edit', { ...shared, prompt, num_images });
@@ -341,12 +341,12 @@ function registerGenerateTools(server, client, options = {}) {
       });
 
       // Multi-source compositing or DNA-anchored edits routinely exceed 120s
-      // server-side. Extend the polling window in those cases to avoid forcing
-      // every call into the timeout-and-recover path via get_generation_status.
-      const heavy = (source_images && source_images.length > 1) || (visual_dna_ids && visual_dna_ids.length > 0);
+      // server-side. But we still need to return before the 180s host cutoff,
+      // so use 150s for all cases. The LLM can call get_generation_status
+      // with wait=true to continue polling.
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 3) * 1000,
-        timeout: heavy ? 240000 : 120000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -576,7 +576,7 @@ function registerGenerateTools(server, client, options = {}) {
           status_args: { generation_ids: batch.ids, wait: true },
           reference_images
         });
-        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 8) * 1000, timeout: 900000 }, 'generate_video');
+        return pollBatch(client, batch, { interval: (batch.ok[0].gen.poll_interval_hint || 8) * 1000, timeout: 150000 }, 'generate_video');
       }
 
       const gen = await client.post('/v1/generate/video', { ...shared, prompt });
@@ -592,7 +592,7 @@ function registerGenerateTools(server, client, options = {}) {
       // 5-min window forced routine calls into the timeout-and-recover path.
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 8) * 1000,
-        timeout: 900000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -678,7 +678,7 @@ function registerGenerateTools(server, client, options = {}) {
       // Veo 3.1, Hailuo 2.3, Seedance 2 at higher durations/resolutions).
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 8) * 1000,
-        timeout: 900000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -746,7 +746,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 8) * 1000,
-        timeout: 300000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1334,7 +1334,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 8) * 1000,
-        timeout: 600000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1435,7 +1435,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 8) * 1000,
-        timeout: 300000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1553,7 +1553,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 8) * 1000,
-        timeout: 600000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1684,7 +1684,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 8) * 1000,
-        timeout: 600000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1770,7 +1770,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 5) * 1000,
-        timeout: 1800000 // 30 minutes — long podcasts are a thing
+        timeout: 150000 // Return before host cutoff; LLM can call get_generation_status with wait=true for long podcasts
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1835,7 +1835,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, startResponse.generation_id, {
         interval: (startResponse.poll_interval_hint || 8) * 1000,
-        timeout: 900000 // 15 minutes — 3D generation is slow
+        timeout: 150000 // Return before host cutoff; LLM can call get_generation_status with wait=true for slow 3D jobs
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -1996,7 +1996,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 5) * 1000,
-        timeout: 300000 // 5 min — split_upscale and multi_shot can take longer
+        timeout: 150000 // Return before host cutoff; LLM can call get_generation_status with wait=true for split_upscale and multi_shot
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;
@@ -2165,7 +2165,7 @@ function registerGenerateTools(server, client, options = {}) {
 
       const poll = await pollOrTimedOut(client, gen.generation_id, {
         interval: (gen.poll_interval_hint || 8) * 1000,
-        timeout: 600000
+        timeout: 150000
       });
       if (poll.timedOut) return poll.timedOut;
       const result = poll.result;