@chrischall/gemini-mcp 1.13.0 → 1.14.1

package/dist/tools/video.js

@@ -3,13 +3,20 @@ import { McpToolError } from '@chrischall/mcp-utils';
 import { slugify, baseName } from '../images.js';
 import { resolveImageInputs } from '../inputs.js';
 import { DEFAULT_VIDEO_MODEL } from '../models.js';
-import { emitMedia, reportShape, resolveAspectRatio, orientationSchema, timeoutMsSchema, idempotencyKeySchema, asyncSchema, maxWaitMsSchema, withProgressHeartbeat, assertLocalInputsAvailable, imagesUrlSchema, imagesFileUrisSchema } from './shared.js';
+import { emitMedia, reportShape, resolveAspectRatio, orientationSchema, timeoutMsSchema, idempotencyKeySchema, asyncSchema, maxWaitMsSchema, withProgressHeartbeat, assertLocalInputsAvailable, imagesUrlSchema, imagesFileUrisSchema, imagesBase64Schema } from './shared.js';
 import { fingerprintRequest } from '../jobs.js';
 import { attachCost } from '../pricing.js';
 import { previewLocalInputsUnlessConfirmed, schemaConfirm } from './_confirm.js';
 /** omni's own aspect-ratio enum — NOT the image tools' ASPECT_RATIOS. */
 const VIDEO_ASPECT_RATIOS = ['16:9', '9:16'];
-const VIDEO_TASKS = ['text_to_video', 'image_to_video', 'reference_to_video', 'edit'];
+const VIDEO_TASKS = ['text_to_video', 'image_to_video', 'reference_to_video', 'edit', 'extend'];
+/**
+ * omni's output resolutions (GA, verified live 2026-09-09). Unlike `delivery`
+ * this is content and not transport: it is the cost lever on the whole video
+ * path, since omni bills video output per token and a 10s clip came back at
+ * ~$0.34 in 360p against roughly triple that in 720p.
+ */
+const VIDEO_RESOLUTIONS = ['360p', '720p', '1080p', '4k'];
 // `delivery` is transport, not content: uri hands back a Files API link the
 // client downloads with the api key, inline returns base64 capped at ~4MB.
 // Left unset here so the client's verified default (uri) applies; this is an
@@ -24,20 +31,23 @@ const VIDEO_DELIVERY = ['inline', 'uri'];
 // "last video" and a "last image" are distinct chains. See src/session.ts.
 export function registerVideoTools(server, client) {
     server.registerTool('gemini_video_generate', {
-        description: 'Generate a short video via the Gemini omni model (preview): text→video, image→video / reference→video ' +
-            '(supply reference image[s]), or edit a prior video (task: "edit" + previous_interaction_id / continue_last). ' +
-            'Output is written to disk as MP4 (video has no inline MCP block). Video runs long — use `async: true` to get a ' +
-            'job_id immediately and poll gemini_get_result, or raise `timeout_ms`. Preview model: needs a funded account.',
+        description: 'Generate a short video (~10s) via the Gemini omni model: text→video, image→video / reference→video ' +
+            '(supply reference image[s]), interpolate between two stills (pass first frame then last frame as images), ' +
+            'or continue a prior video (task: "edit" or "extend" + previous_interaction_id / continue_last; extensions ' +
+            'add ~3-10s each, to ~40s total). Cost scales with `resolution` — draft at 360p, keep at 1080p/4k. ' +
+            'Output is written to disk as MP4 (video has no inline MCP block). Video runs long — give it a `max_wait_ms` budget ' +
+            '(or `async: true` + gemini_get_result on a local install), or raise `timeout_ms`. Needs a funded account.',
         annotations: { readOnlyHint: false, openWorldHint: true },
         inputSchema: {
             prompt: z.string().min(1).describe('Description of the video to generate (or the edit instruction when task=edit)'),
             aspect_ratio: z.enum(VIDEO_ASPECT_RATIOS).optional().describe('Exact output aspect ratio (omni: 16:9 or 9:16). `orientation` is the plain-language shorthand; this wins if both are given.'),
             orientation: orientationSchema,
-            task: z.enum(VIDEO_TASKS).optional().describe('text_to_video (default), image_to_video / reference_to_video (need image input), or edit (needs previous_interaction_id)'),
+            task: z.enum(VIDEO_TASKS).optional().describe('text_to_video (default), image_to_video / reference_to_video (need image input), or edit / extend (need previous_interaction_id)'),
+            resolution: z.enum(VIDEO_RESOLUTIONS).optional().describe('Output resolution (default 720p). Video is billed per output token, so 360p costs roughly a third of 720p — use it for drafts'),
             images: z.array(z.string().min(1)).optional().describe('Reference image path(s) for image_to_video / reference_to_video'),
             images_url: imagesUrlSchema('Reference stills'),
             images_file_uris: imagesFileUrisSchema('Reference stills'),
-            images_base64: z.array(z.string().min(1)).optional().describe('Reference images as base64 strings or data URIs. Last resort: prefer images_url or images_file_uris, which keep image bytes out of the conversation'),
+            images_base64: imagesBase64Schema(),
             from_clipboard: z.boolean().optional().describe('Use the image currently on the macOS clipboard as a reference'),
             filename: z.string().optional().describe('Base filename for the output video (extension stripped; default: slugified prompt)'),
             output_dir: z.string().optional().describe('Directory to write the video to (default: $GEMINI_OUTPUT_DIR or cwd)'),
@@ -78,7 +88,9 @@ export function registerVideoTools(server, client) {
         // fingerprint so "portrait" and "9:16" dedup to one billable job.
         const aspectRatio = resolveAspectRatio(args, VIDEO_ASPECT_RATIOS);
         const fingerprint = fingerprintRequest('gemini_video_generate', {
-            model, prompt: args.prompt, aspect_ratio: aspectRatio, task: args.task,
+            // `resolution` is IN the fingerprint (a 4k render is not the 360p one)
+            // where `delivery` is not (same bytes, different transport).
+            model, prompt: args.prompt, aspect_ratio: aspectRatio, resolution: args.resolution, task: args.task,
             images: args.images, images_base64: args.images_base64, from_clipboard: args.from_clipboard,
             images_url: args.images_url, images_file_uris: args.images_file_uris,
             previous_interaction_id: previousInteractionId,
@@ -90,6 +102,7 @@ export function registerVideoTools(server, client) {
                 images: inputs.length ? inputs : undefined,
                 model: args.model,
                 aspectRatio,
+                resolution: args.resolution,
                 task: args.task,
                 previousInteractionId,
                 timeoutMs: args.timeout_ms,
@@ -104,6 +117,10 @@ export function registerVideoTools(server, client) {
             // The RESOLVED ratio: omni takes only 16:9/9:16, so a caller who asked
             // in words can see which of the two their request became.
             const meta = reportShape({ model, interaction_id: r.id }, { ...args, aspect_ratio: aspectRatio });
+            // Echoed for the same reason the resolved ratio is: it changed what
+            // the caller was billed, so it has to be visible in the result.
+            if (args.resolution)
+                meta.resolution = args.resolution;
             if (previousInteractionId)
                 meta.previous_interaction_id = previousInteractionId;
             if (r.usage)
@@ -119,7 +136,7 @@ export function registerVideoTools(server, client) {
                 media,
                 base: r.videos.length > 1 ? `${slug}-${String(i + 1).padStart(2, '0')}` : slug,
             }));
-            return emitMedia(named, 'video', { output_dir: args.output_dir, sink: client.mediaSink }, meta);
+            return emitMedia(named, 'video', { output_dir: args.output_dir, sink: client.mediaSink, sidecar: { prompt: args.prompt } }, meta);
         });
     });
 }

package/dist/upload-url.js

@@ -5,9 +5,9 @@
  * The problem these solve is the mirror image of `/media`: a signed GET lets a
  * shell *download* generated media with no auth header, but uploading a
  * reference photo still required `images_base64` (~14k tokens per JPEG, silently
- * corrupting on a truncated read) or `POST /upload` (which 401s from any shell
- * that doesn't hold the MCP session's bearer token). A signed PUT URL closes
- * that gap: the authenticated MCP session mints it, and then a bare
+ * corrupting on a truncated read) or the Worker's `POST /upload`, which 401'd
+ * from any shell without the MCP session's bearer token and is gone entirely
+ * now that the Worker is. A signed PUT URL closes that gap: the authenticated MCP session mints it, and then a bare
  * `curl -X PUT --data-binary @photo.jpg "<url>"` stores the bytes.
  *
  * Same HMAC secret as the media GET URLs, but a DIFFERENT payload shape, so the

package/dist/version.js

@@ -1,2 +1,2 @@
 /** Single source of the server version. release-please bumps the literal below. */
-export const VERSION = '1.13.0'; // x-release-please-version
+export const VERSION = '1.14.1'; // x-release-please-version

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chrischall/gemini-mcp",
-  "version": "1.13.0",
+  "version": "1.14.1",
   "mcpName": "io.github.chrischall/gemini-mcp",
   "description": "Gemini image-generation MCP server for Claude — developed and maintained by AI (Claude Code)",
   "author": "Claude Code (AI) <https://www.anthropic.com/claude>",
@@ -46,16 +46,16 @@
     "typecheck": "tsc -p tsconfig.test.json"
   },
   "dependencies": {
-    "@chrischall/mcp-utils": "^0.23.0",
-    "@modelcontextprotocol/sdk": "^1.29.0",
+    "@chrischall/mcp-utils": "^0.26.1",
+    "@modelcontextprotocol/sdk": "^1.30.0",
     "dotenv": "^17.4.0",
-    "zod": "^4.4.2"
+    "zod": "^4.5.4"
   },
   "devDependencies": {
     "@types/node": "^26.0.0",
-    "@vitest/coverage-v8": "^4.1.2",
+    "@vitest/coverage-v8": "^5.0.0",
     "esbuild": "^0.28.0",
     "typescript": "^7.0.2",
-    "vitest": "^4.1.2"
+    "vitest": "^5.0.0"
   }
 }

package/server.json

@@ -6,12 +6,12 @@
     "url": "https://github.com/chrischall/gemini-mcp",
     "source": "github"
   },
-  "version": "1.13.0",
+  "version": "1.14.1",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@chrischall/gemini-mcp",
-      "version": "1.13.0",
+      "version": "1.14.1",
       "transport": {
         "type": "stdio"
       },

package/skills/gemini-mcp/SKILL.md

@@ -118,6 +118,9 @@ not an error.
   any number of calls until then.
 - On stdio, an `images` path referenced **more than once in a session** is auto-uploaded to the
   Files API (keyed on path + mtime + size) so the bytes stop being re-sent.
+- `images_base64` is uploaded on the **first** sighting, not the second — the tokens are already
+  spent by then. The result reports it under `image_inputs.base64_uploaded[].file_uri`: pass that
+  to `images_file_uris` on the next call instead of pasting the bytes again.
 
 ### Files API
 | Tool | Description |
@@ -145,8 +148,8 @@ curl -X POST https://<hosted deployment>/upload \
 ### Video & Music (preview — funded account)
 | Tool | Description |
 |------|-------------|
-| `gemini_video_generate(prompt, aspect_ratio?, task?, images_url?, images_file_uris?, images?, images_base64?, from_clipboard?, previous_interaction_id?, continue_last?, model?, filename?, output_dir?, timeout_ms?, idempotency_key?, async?)` | Generate a short video via the Gemini omni model: `text_to_video` (default), `image_to_video` / `reference_to_video` (supply reference image[s]), or `edit` (with `previous_interaction_id` / `continue_last`). `aspect_ratio` is `16:9` or `9:16`. Written to disk as **MP4**. Runs long — use `async: true` + `gemini_get_result`, or raise `timeout_ms`. |
-| `gemini_music_generate(prompt, model?, audio_format?, images_url?, images_file_uris?, images?, images_base64?, from_clipboard?, previous_interaction_id?, continue_last?, filename?, output_dir?, inline?, timeout_ms?, idempotency_key?, async?)` | Generate music from a prompt (mood/genre/instruments/structure/lyrics) via a Lyria model: `lyria-3-clip-preview` (~30s, default) or `lyria-3-pro-preview` (longer, WAV-capable). `audio_format` `mp3` (default) or `wav` (**Pro-only**). Written to disk as MP3/WAV, or returned inline. |
+| `gemini_video_generate(prompt, aspect_ratio?, resolution?, task?, images_url?, images_file_uris?, images?, images_base64?, from_clipboard?, previous_interaction_id?, continue_last?, model?, filename?, output_dir?, timeout_ms?, idempotency_key?, async?)` | Generate a short video (~10s) via the Gemini omni model: `text_to_video` (default), `image_to_video` / `reference_to_video` (supply reference image[s]), interpolation (pass first frame then last frame as `images`), or `edit` / `extend` (with `previous_interaction_id` / `continue_last`; extensions add ~3-10s each, to ~40s). `aspect_ratio` is `16:9` or `9:16`; `resolution` is `360p`/`720p` (default)/`1080p`/`4k` and is the cost lever — a 10s 360p clip runs about a third of 720p. Written to disk as **MP4**. Runs long — use `async: true` + `gemini_get_result`, or raise `timeout_ms`. |
+| `gemini_music_generate(prompt, model?, images_url?, images_file_uris?, images?, images_base64?, from_clipboard?, filename?, output_dir?, inline?, timeout_ms?, idempotency_key?, async?)` | Generate music from a prompt (mood/genre/instruments/structure/lyrics) via a Lyria model: `lyria-3-clip-preview` (30s instrumental, default, $0.04), `lyria-3.5` (full-length song with vocals, $0.08) or `lyria-3-pro-preview` ($0.08). Output is MP3. **Single-turn** — there is no follow-up refinement, so put the whole brief in the prompt. Written to disk, or returned inline. |
 
 ### Async / idempotency (any generation tool)
 | Tool | Description |
@@ -217,7 +220,8 @@ gemini_image_edit(
 → returns path to the edited image
 ```
 `images_base64` is for bytes you actually have — a file you `Read`/encode, a URL
-you fetch, or a `data:` URI the user pastes as **text**.
+you fetch, or a `data:` URI the user pastes as **text**. Send them once: the result's
+`image_inputs.base64_uploaded[].file_uri` is a `files/<id>` to reuse via `images_file_uris`.
 
 **Iterate on ONE image conversationally (multi-turn):**
 ```
@@ -231,21 +235,25 @@ r3 = gemini_interact(input: "warmer lighting", continue_last: true)
 ```
 Prefer this over re-running `gemini_image_edit` when you're making a *series* of incremental edits — the model keeps the prior result in context. Every result echoes `interaction_id` (and `previous_interaction_id` when chaining) plus a `hint` with the exact follow-up call.
 
-**Generate a video (preview — runs long, use async):**
+**Generate a video (runs long, use async):**
 ```
 job = gemini_video_generate(prompt: "a paper boat sailing down a rain gutter, cinematic",
-                            aspect_ratio: "16:9", async: true)
+                            aspect_ratio: "16:9", resolution: "360p", async: true)
    → { job_id, status: "running" }   (returns immediately — no host timeout)
 gemini_get_result(job_id: job.job_id)
    → "running" until done, then the MP4 path on disk
+# Draft at 360p, re-run the keeper at 1080p — video bills per output token, so the
+# resolution is roughly the price.
 # Animate a still instead: gemini_video_generate(prompt: "…", task: "image_to_video", images: ["/path/still.png"])
+# Interpolate between two stills: images: ["/path/first.png", "/path/last.png"]
+# Extend a clip: gemini_video_generate(prompt: "…", task: "extend", continue_last: true)
 ```
 
-**Generate music (preview):**
+**Generate music:**
 ```
 gemini_music_generate(prompt: "warm lo-fi hip hop, mellow Rhodes, vinyl crackle, 70bpm")
-   → ~30s MP3 on disk (lyria-3-clip-preview)
-# Longer / WAV: gemini_music_generate(prompt: "…", model: "lyria-3-pro-preview", audio_format: "wav")
+   → 30s MP3 on disk (lyria-3-clip-preview)
+# Full-length song with vocals: gemini_music_generate(prompt: "…", model: "lyria-3.5")
 ```
 **⚠️ Chat-pasted/attached images can't be fed to these tools directly.** A pasted
 image reaches the assistant as a *vision* block — the assistant can SEE it but