@viralspin/mcp 0.1.0 → 0.2.2

package/dist/tools/ideas.js

@@ -0,0 +1,158 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.listIdeasTool = exports.queueIdeaTool = void 0;
+/**
+ * queue_idea / list_ideas — the Prompt/Idea History bridge. Claude brainstorms
+ * with the user, then queues each idea (hook + script + an asset plan) here; it
+ * persists in Supabase (via /v1/spin/ideas) and shows up in the user's web
+ * /spin/history, cross-session and cross-browser. The asset_plan tells the user
+ * what's needed to execute (existing clips, AI B-roll, web images, kinetic text,
+ * or footage to film).
+ */
+const client_js_1 = require("../client.js");
+exports.queueIdeaTool = {
+    name: "queue_idea",
+    description: "Queue a content idea into the user's persistent Prompt/Idea History (shows in their web /spin/history, " +
+        "cross-browser). Use after brainstorming or after analyzing a saved/pasted post the user wants to replicate. " +
+        "Give it a `hook`, a ready-to-film `script` (they read it off the teleprompter at /create), the `topic`, and " +
+        "an `asset_plan` — per shot, say what to use: existing clip / AI-generated B-roll / web image (logo, person) / " +
+        "kinetic text overlay / footage to film. Set `source_url` if it came from a saved post. " +
+        "For a punchy 5-7s relatable joke where the VALUE IS THE OVERLAY CAPTION (not narration) — e.g. a video of " +
+        "someone on a plane captioned 'Flight attendant: OMG is there anyone onboard who can align cross-functional " +
+        "stakeholders while navigating ambiguity?' — use format:'reel_7s' and put the line in `caption`. " +
+        "GROUND every reel_7s in what is CURRENTLY GOING VIRAL in the creator's niche: call get_creator_profile " +
+        "for their niche + handles + topics, research the caption-reel formats + in-group jokes trending in that " +
+        "niche right now (their and adjacent creators' recent hits), and adapt a PROVEN pattern to their audience — " +
+        "don't invent a random joke. When generating a WEEK's batch of ideas, include AT LEAST ONE reel_7s modeled " +
+        "on that week's viral trend in the niche (set `source_url` to the post it's modeled on when there is one). " +
+        "reel_7s ASSET PLAN — if the caption has a SETUP → PAYOFF shape (before/after, question → reaction, 'four " +
+        "hours later:'), SPLIT the asset_plan into ONE BEAT PER SEGMENT so the shoot app can guide the creator " +
+        "part-by-part; a single-line caption = a single beat. Give each segment its own overlay line QUOTED in " +
+        "`note` (e.g. note: 'Overlay: \"Four hours later:\". Same framing as segment 1 - the contrast IS the joke.'), " +
+        "its own `search_terms`, and kind:'existing' — so the app can AI-pick that segment from the creator's camera " +
+        "roll (a 7s clip OR a photo) and they can still film/overwrite it themselves. " +
+        "HOOK PATTERN — a strong attention-grab is to OPEN ON A SURPRISED REACTION, then demonstrate the payoff " +
+        "('wait, what?!' → here's the thing). For ideas that PROMOTE VIRALSPIN (which are encouraged), lead the " +
+        "asset_plan with a reaction beat: first shot kind:'film' note the creator's surprised/delighted reaction " +
+        "(e.g. 'reacts stunned to how fast the reel was made'), THEN the demonstration beats. The mobile shoot flow " +
+        "(Take → React → Show → Build) mirrors this, so a react-first plan maps cleanly to how it's executed. " +
+        "SHOT PHRASING — write each `shot` as a SHORT, PLAIN, second-person creator ACTION they can just DO, not a " +
+        "cinematic third-person screenplay direction. GOOD: 'Show/gesture a perfect prompt', 'Sift through a bunch of AI " +
+        "output and pick a few good ones', 'React impressed at the polished result'. BAD (too abstract/wordy): 'Creator " +
+        "looking thoughtful, gesturing beyond the camera to a conceptual future', 'Hands sifting through a grid of " +
+        "AI-generated text/images on a tablet, selecting a few'. If a beat is really an EMOTIONAL/impressed MOMENT (e.g. " +
+        "'confidently presenting a refined high-quality result'), make it a REACTION the creator performs to camera — an " +
+        "impressed facial reaction — NOT a staged 'show' of holding up a screen. " +
+        "TAKE vs SHOW — the creator ON CAMERA (talking, gesturing, reacting) belongs to the Take + React steps, so those " +
+        "are kind:'film'. The 'SHOW IT' beats are footage of the SUBJECT — a screen recording, a demo, product/B-roll — " +
+        "NOT the creator; make them kind:'existing' (a clip the user records/provides) or kind:'ai_broll', and describe " +
+        "WHAT'S ON SCREEN, not the person. NEVER write a show/demo beat as 'creator looking thoughtful' or 'creator " +
+        "pointing at a screen' — that's a talking-head (Take). E.g. a demo beat = existing 'Screen-record the bad-vs-good " +
+        "prompt output', not film 'Creator points to the screen'. " +
+        "NO REDUNDANT TAKE BEATS — the creator reads the WHOLE script off the teleprompter in step 1, so NEVER add a " +
+        "kind:'film' beat that is just 'talk/gesture to camera as you tee up the idea': that IS the Take, and it orphans " +
+        "in the shoot UI. A kind:'film' beat must be a PERFORMED MOMENT — a face or a GESTURE — because those become " +
+        "real tasks in the React step (step 2). Reach for a gesture when a phrase invites one: 'the real secret…' → film " +
+        "'Shhh - index finger to your lips as you tease the secret'. " +
+        "EVERY LINE SHOULD HAVE SOMETHING ON SCREEN — the shoot app shows the script beside an 'on screen' column, so " +
+        "cover the script with beats, and lean on the NON-FILM kinds so the creator isn't asked to shoot everything: " +
+        "kind:'web_image' = a logo/graphic/image PULLED FROM THE WEB, floated in a card beside the creator's head (use it " +
+        "when a concept has an obvious visual — 'AI success' → a trending-up chart / trophy); kind:'kinetic' = text " +
+        "overlays AND animated motion graphics. Describe a motion graphic in PLAIN TERMS of what the viewer sees ('a " +
+        "terminal window that types the prompt out') — never name an implementation (don't say 'SVG', 'Lottie' or " +
+        "'Remotion'): these are drawn natively on the phone, so an implementation name is both wrong and meaningless to " +
+        "the creator reading the plan. A terminal/typing animation is the one bespoke motion graphic that renders today " +
+        "(say 'terminal' or 'typing' in the shot and it's picked up); any other kinetic beat renders as its text overlay, " +
+        "so put the words you want burned on screen IN QUOTES. " +
+        "For any web_image/kinetic overlay DON'T pin an exact position — say it floats above the creator's head " +
+        "or wherever there's space, and note that placement is resolved at render so it won't collide with the captions.",
+    inputSchema: {
+        type: "object",
+        properties: {
+            hook: { type: "string", description: "The scroll-stopping opening line / concept." },
+            script: { type: "string", description: "The full narration script to film (teleprompter-ready)." },
+            topic: { type: "string", description: "The core topic / pillar this idea belongs to." },
+            tags: { type: "array", items: { type: "string" }, description: "Optional tags." },
+            asset_plan: {
+                type: "array",
+                description: "Per-shot plan: [{shot, kind:'existing'|'ai_broll'|'web_image'|'kinetic'|'film', note, narration, search_terms, asset_url?}]. " +
+                    "ALWAYS set `search_terms` on any SHOW beat (existing / ai_broll) AND on every `web_image` beat — on a " +
+                    "web_image it's the WEB SEARCH we run to find the graphic, and a result is only kept if its title mentions " +
+                    "one of these terms, so an abstract concept with no concrete nouns finds nothing and the card falls back to " +
+                    "plain text (e.g. 'AI success' → search_terms 'trophy, winner podium, chart going up, success'). " +
+                    "Set `asset_url` instead when a SPECIFIC image is wanted; then no search runs. search_terms are " +
+                    "CONCRETE VISUAL NOUNS, comma-separated, " +
+                    "that could literally appear in a photo — the shoot app matches these against on-device Vision labels of " +
+                    "the creator's camera roll. `shot` is written for a HUMAN to read and is often abstract, which matches " +
+                    "nothing; `search_terms` is what the matcher actually searches. E.g. shot 'Sift through a bunch of AI " +
+                    "output and pick a few good ones' → search_terms 'tablet, laptop screen, computer monitor, grid of images, " +
+                    "hands typing'. Use plain everyday objects/places/people (laptop, screen, desk, phone, coffee, office, " +
+                    "whiteboard) — NOT abstractions ('AI output', 'workflow', 'the future'). " +
+                    "ALWAYS set `narration` to the EXACT word(s) from `script` the creator is SPEAKING while this beat is on " +
+                    "screen — the beats run in order and together cover the whole script, so each beat's narration is its slice of " +
+                    "the script verbatim. A slice does NOT have to be a whole sentence: when an overlay is triggered by SPECIFIC " +
+                    "WORDS, CUT THE SLICE AT THOSE WORDS so the beat lands on the row it belongs to. E.g. for 'The real secret to " +
+                    "AI success isn't learning to prompt.' with a shush gesture on 'the real secret' and a web graphic on 'AI " +
+                    "success', emit TWO beats — narration 'The real secret' then 'to AI success isn't learning to prompt.' — " +
+                    "NOT one sentence-beat plus a graphic pushed onto the NEXT sentence, which mis-times the graphic. This is what the shoot UI shows in quotes under each shot, so the creator knows what " +
+                    "they're saying during it. (A kinetic text-overlay beat still has narration playing under it.)",
+                items: {
+                    type: "object",
+                    properties: {
+                        shot: { type: "string" },
+                        kind: { type: "string", enum: ["existing", "ai_broll", "web_image", "kinetic", "film"] },
+                        note: { type: "string" },
+                        narration: { type: "string", description: "The verbatim script line(s) spoken during this beat." },
+                        search_terms: { type: "string", description: "Concrete visual nouns (comma-separated) to match the creator's camera roll, e.g. 'tablet, laptop screen, grid of images'." },
+                        asset_url: { type: "string" },
+                    },
+                    required: ["shot", "kind"],
+                },
+            },
+            source_url: { type: "string", description: "The saved/pasted post URL this idea is modeled on, if any." },
+            format: { type: "string", enum: ["video", "carousel", "image", "reel_7s"], description: "How it's executed: 'video' (film off the teleprompter) | 'carousel' (image carousel) | 'image' | 'reel_7s' (a 5-7s caption reel — see below). Default video." },
+            caption: { type: "string", description: "For format 'reel_7s' ONLY: the single punchy OVERLAY CAPTION that IS the joke/value (e.g. \"Flight attendant: OMG is there anyone onboard who can align cross-functional stakeholders while navigating ambiguity?\"). The user films 5-7s of themselves reacting / doing a light action (or any B-roll — it barely matters); the caption carries it. Put the caption here AND in `hook`." },
+            week: { type: "string", description: "ISO week tag for weekly batches, e.g. '2026-W27'. Group a week's 5 ideas under the same value." },
+        },
+    },
+    async execute(args) {
+        // For a 7s caption reel the overlay caption IS the content — use it as the hook/script if not given.
+        const hook = args.hook ?? args.caption;
+        const script = args.script ?? (args.format === "reel_7s" ? args.caption : undefined);
+        if (!hook && !script)
+            return "Give the idea at least a `hook` or a `script` (or a `caption` for reel_7s).";
+        const r = await (0, client_js_1.post)("/v1/spin/ideas", {
+            hook,
+            script,
+            topic: args.topic,
+            tags: args.tags,
+            asset_plan: args.asset_plan,
+            source_url: args.source_url,
+            format: args.format,
+            caption: args.caption,
+            week: args.week,
+        });
+        const planNote = args.asset_plan?.length ? ` with a ${args.asset_plan.length}-shot asset plan` : "";
+        return (`Queued idea ${r.id}${planNote} → it's in the user's Prompt/Idea History (/spin/history), saved across sessions + browsers.\n` +
+            `They can open it, film the script on the teleprompter at /create, and the plan says what B-roll to use vs film.`);
+    },
+};
+exports.listIdeasTool = {
+    name: "list_ideas",
+    description: "List the user's queued content ideas (their Prompt/Idea History). Optionally filter by status " +
+        "(queued / scripted / filmed / done / dismissed). Use to resume planning or pick the next idea to execute.",
+    inputSchema: {
+        type: "object",
+        properties: {
+            status: { type: "string", enum: ["queued", "scripted", "filmed", "done", "dismissed"], description: "Optional status filter." },
+        },
+    },
+    async execute(args) {
+        const q = args.status ? `?status=${encodeURIComponent(args.status)}` : "";
+        const r = await (0, client_js_1.get)(`/v1/spin/ideas${q}`);
+        if (!r.ideas.length)
+            return args.status ? `No ideas with status "${args.status}".` : "No ideas queued yet — brainstorm some and queue_idea them.";
+        const lines = r.ideas.map((i) => `- [${i.status}] ${i.hook || "(no hook)"}${i.topic ? ` · ${i.topic}` : ""}${i.source_url ? ` · from ${i.source_url}` : ""}  (id ${i.id})`);
+        return [`${r.ideas.length} idea(s) in the Prompt/Idea History:`, ...lines].join("\n");
+    },
+};

package/dist/tools/image-ads.js

@@ -0,0 +1,156 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.publishToMetaTool = exports.assembleImageAdTool = exports.prepareProductAssetTool = void 0;
+const client_js_1 = require("../client.js");
+const APP_BASE = process.env.VIRALSPIN_APP_URL || "https://app.viralspin.ai";
+exports.prepareProductAssetTool = {
+    name: "prepare_product_asset",
+    description: "Download and verify a product image from a local path or public URL, upload it to ViralSpin " +
+        "cloud storage, and return a stable cloud URL ready for Higgsfield image ad generation.",
+    inputSchema: {
+        type: "object",
+        required: ["image_url"],
+        properties: {
+            image_url: {
+                type: "string",
+                description: "Local file path (e.g. C:/photos/product.jpg) or public URL of the product image.",
+            },
+        },
+    },
+    async execute({ image_url }) {
+        if (image_url.startsWith("http://") || image_url.startsWith("https://")) {
+            return `✅ Asset ready (already a URL):\n\n  ${image_url}\n\nPass this to assemble_image_ad.`;
+        }
+        const { urls } = await (0, client_js_1.uploadFiles)([image_url]);
+        return [
+            `✅ Product asset uploaded:`,
+            ``,
+            `  ${urls[0]}`,
+            ``,
+            `Pass this URL to assemble_image_ad.`,
+        ].join("\n");
+    },
+};
+exports.assembleImageAdTool = {
+    name: "assemble_image_ad",
+    description: "Generate image ad variants from a product photo using Higgsfield AI. " +
+        "Uploads the product image, generates styled variants with your prompt, then creates a " +
+        "client review link showing all variants in a carousel. " +
+        "Requires Higgsfield credentials saved in ViralSpin Settings.",
+    inputSchema: {
+        type: "object",
+        required: ["product_image_path", "prompt"],
+        properties: {
+            product_image_path: {
+                type: "string",
+                description: "Local file path (e.g. C:/photos/product.jpg) or public URL of the product image.",
+            },
+            prompt: {
+                type: "string",
+                description: "Higgsfield styling prompt describing the desired ad aesthetic. " +
+                    "Example: 'luxury product shot on white marble, soft studio lighting, premium brand aesthetic'",
+            },
+            pin: {
+                type: "string",
+                description: "Optional 4-digit PIN to protect the client review link.",
+            },
+        },
+    },
+    async execute({ product_image_path, prompt, pin }) {
+        // Upload local files; pass through URLs unchanged
+        let imageUrl = product_image_path;
+        if (!product_image_path.startsWith("http://") && !product_image_path.startsWith("https://")) {
+            const { urls } = await (0, client_js_1.uploadFiles)([product_image_path]);
+            imageUrl = urls[0];
+        }
+        // Start async generation job — Higgsfield lives on Python backend
+        const { job_id } = await (0, client_js_1.pyPost)("/v1/higgsfield/generate", {
+            product_image_url: imageUrl,
+            prompt,
+        });
+        // Poll until done (~3 min timeout)
+        let variants = [];
+        for (let attempt = 0; attempt < 40; attempt++) {
+            await new Promise((r) => setTimeout(r, 5000));
+            const status = await (0, client_js_1.pyGet)(`/v1/higgsfield/status/${job_id}`);
+            if (status.status === "complete") {
+                variants = status.variant_urls ?? [];
+                break;
+            }
+            if (status.status === "failed") {
+                throw new Error(`Higgsfield generation failed: ${status.error ?? "unknown error"}`);
+            }
+        }
+        if (!variants.length) {
+            throw new Error("Generation timed out after ~3 minutes — try again or check Higgsfield credentials");
+        }
+        // Create a review session with image variants displayed as a carousel
+        const { token } = await (0, client_js_1.post)("/v1/review-sessions", {
+            output_path: variants[0],
+            metadata: {
+                media_type: "image",
+                variants,
+                caption: prompt,
+                platform_targets: ["instagram", "facebook"],
+            },
+            pin: pin ?? null,
+            expires_hours: 168,
+        });
+        const reviewUrl = `${APP_BASE}/review/${token}`;
+        return [
+            `✅ ${variants.length} image ad variant${variants.length > 1 ? "s" : ""} generated!`,
+            ``,
+            `Review link: ${reviewUrl}`,
+            ``,
+            `Send this to your client — they'll see all ${variants.length} variants as a carousel`,
+            `and can Approve or Request Changes.`,
+            pin ? `PIN: ${pin}` : ``,
+            ``,
+            `To publish an approved variant:`,
+            `  publish_to_meta with media_url="${variants[0]}" and your caption.`,
+        ]
+            .filter((l) => l !== undefined)
+            .join("\n");
+    },
+};
+exports.publishToMetaTool = {
+    name: "publish_to_meta",
+    description: "Publish an image (or video) to Instagram and/or Facebook. " +
+        "Use after a client approves an image ad variant from assemble_image_ad.",
+    inputSchema: {
+        type: "object",
+        required: ["media_url", "platforms", "caption"],
+        properties: {
+            media_url: {
+                type: "string",
+                description: "Public URL of the approved image or video to publish.",
+            },
+            platforms: {
+                type: "array",
+                items: { type: "string", enum: ["instagram", "facebook"] },
+                description: "Target platforms: 'instagram', 'facebook', or both.",
+            },
+            caption: {
+                type: "string",
+                description: "Post caption (include hashtags here if desired).",
+            },
+        },
+    },
+    async execute({ media_url, platforms, caption }) {
+        const result = await (0, client_js_1.post)("/v1/publish/schedule", {
+            video_url: media_url,
+            platforms,
+            caption,
+            publish_now: true,
+        });
+        const lines = ["📸 Published to Meta:"];
+        const results = result.results ?? {};
+        for (const [platform, res] of Object.entries(results)) {
+            lines.push(`  ${res.success ? "✅" : "❌"} ${platform}${res.error ? `: ${res.error}` : ""}`);
+        }
+        if (!Object.keys(results).length) {
+            lines.push("  (no platform results returned — check your connected accounts in Settings)");
+        }
+        return lines.join("\n");
+    },
+};

package/dist/tools/kinetic-text.js

@@ -0,0 +1,122 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createKineticTextTool = void 0;
+const child_process_1 = require("child_process");
+const path_1 = __importDefault(require("path"));
+const engine_js_1 = require("../engine.js");
+const EFFECTS = ["typing", "billboard_flip", "isometric"];
+const STYLES = ["terminal", "cinematic", "minimal"];
+const FORMATS = ["9x16", "16x9", "1x1"];
+function runKinetic(args) {
+    return new Promise((resolve, reject) => {
+        const child = (0, child_process_1.spawn)(process.platform === "win32" ? "node.exe" : "node", [path_1.default.join(engine_js_1.ENGINE_DIR, "kinetic.mjs"), ...args], { cwd: engine_js_1.ENGINE_DIR, env: { ...process.env } });
+        let stdout = "";
+        child.stdout.on("data", (d) => { stdout += d.toString(); });
+        child.stderr.on("data", (d) => process.stderr.write(d));
+        child.on("error", reject);
+        child.on("close", (code) => {
+            const m = stdout.match(/RESULT:(.+)/);
+            if (code === 0 && m)
+                resolve(m[1].trim());
+            else
+                reject(new Error(`Kinetic render failed (exit ${code}). ${stdout.slice(-300)}`));
+        });
+    });
+}
+exports.createKineticTextTool = {
+    name: "create_kinetic_text",
+    description: "Render a standalone cinematic typing-animation video — great for explainer intros that visually demo " +
+        "what a user types into the prompt. Text is typed out character-by-character with a blinking cursor " +
+        "while a 3D camera slowly pans, giving a cinematic 'Star Wars pull-back' feel. " +
+        "Three effects: 'typing' (character-by-character with 3D camera), " +
+        "'billboard_flip' (words fly from Z-depth and snap into place), " +
+        "'isometric' (letters extrude upward on a tilted plane). " +
+        "Three styles: 'terminal' (green-on-dark, monospace, scanlines), " +
+        "'cinematic' (white-on-black, serif, vignette), " +
+        "'minimal' (transparent background, for overlay use). " +
+        "Returns the path to a finished MP4 — feed it to publish_now or as A-roll in create_video_assembly.",
+    inputSchema: {
+        type: "object",
+        properties: {
+            text: {
+                type: "string",
+                description: "The text to animate — e.g. the prompt the user would type.",
+            },
+            effect: {
+                type: "string",
+                enum: EFFECTS,
+                description: "Animation style. Default: typing.",
+            },
+            style: {
+                type: "string",
+                enum: STYLES,
+                description: "Visual theme. terminal = green monospace; cinematic = white serif; minimal = transparent BG. Default: terminal.",
+            },
+            format: {
+                type: "string",
+                enum: FORMATS,
+                description: "Output aspect ratio. Default: 9x16 (vertical).",
+            },
+            font_size: {
+                type: "number",
+                description: "Font size in px. Auto-sized by format if omitted (72px for 9x16, 96px for 16x9).",
+            },
+            camera: {
+                type: "object",
+                description: "Override the 3D camera end position. Each field defaults to the effect's cinematic preset.",
+                properties: {
+                    rotateX: { type: "number", description: "X-axis tilt in degrees (positive = tilt top toward viewer)." },
+                    rotateY: { type: "number", description: "Y-axis turn in degrees." },
+                    rotateZ: { type: "number", description: "Z-axis roll in degrees." },
+                    translateZ: { type: "number", description: "Z-depth offset in px (negative = pull back/zoom out)." },
+                },
+            },
+            duration_sec: {
+                type: "number",
+                description: "Override clip length in seconds. Default: auto (≈ text.length × 0.1s + 1.5s hold).",
+            },
+            out: {
+                type: "string",
+                description: "Output MP4 path. Defaults to engine/out/kinetic_<timestamp>.mp4.",
+            },
+        },
+        required: ["text"],
+    },
+    async execute(args) {
+        if (!(0, engine_js_1.engineAvailable)("kinetic.mjs")) {
+            return ("create_kinetic_text needs the ViralSpin engine, which isn't bundled in this install.\n" +
+                "Set VIRALSPIN_ASSEMBLY_DIR to a local engine checkout to enable it.");
+        }
+        const cli = ["--text", args.text];
+        if (args.effect)
+            cli.push("--effect", args.effect);
+        if (args.style)
+            cli.push("--style", args.style);
+        if (args.format)
+            cli.push("--format", args.format);
+        if (args.font_size)
+            cli.push("--font-size", String(args.font_size));
+        if (args.duration_sec)
+            cli.push("--duration-sec", String(args.duration_sec));
+        if (args.out)
+            cli.push("--out", args.out);
+        if (args.camera) {
+            const c = args.camera;
+            if (c.rotateX != null)
+                cli.push("--camera-rx", String(c.rotateX));
+            if (c.rotateY != null)
+                cli.push("--camera-ry", String(c.rotateY));
+            if (c.rotateZ != null)
+                cli.push("--camera-rz", String(c.rotateZ));
+            if (c.translateZ != null)
+                cli.push("--camera-tz", String(c.translateZ));
+        }
+        const outPath = await runKinetic(cli);
+        return (`Kinetic text render complete.\n` +
+            `Output: ${outPath}\n` +
+            `Use this as a hook intro by passing it as A-roll to create_video_assembly, or publish directly with publish_now.`);
+    },
+};

package/dist/tools/local-index.js

@@ -0,0 +1,84 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.searchLocalIndexTool = exports.syncLocalIndexTool = void 0;
+/**
+ * sync_local_index / search_local_index — make the user's LOCAL footage count in
+ * cloud idea-planning. sync_local_index analyzes a folder on the GPU (free when
+ * local mode is on) and pushes TAGS ONLY to the cloud catalog (files stay local);
+ * search_local_index finds existing local clips that fit a script/idea so Claude
+ * can recommend reusing them instead of filming new B-roll.
+ */
+const child_process_1 = require("child_process");
+const path_1 = __importDefault(require("path"));
+const engine_js_1 = require("../engine.js");
+const client_js_1 = require("../client.js");
+function runIndex(folder) {
+    return new Promise((resolve, reject) => {
+        const child = (0, child_process_1.spawn)(process.platform === "win32" ? "node.exe" : "node", [path_1.default.join(engine_js_1.ENGINE_DIR, "index-folder.mjs"), folder], { cwd: engine_js_1.ENGINE_DIR, env: { ...process.env } });
+        let out = "";
+        child.stdout.on("data", (d) => { out += d.toString(); });
+        child.stderr.on("data", (d) => process.stderr.write(d));
+        child.on("error", reject);
+        child.on("close", (code) => {
+            const m = out.match(/RESULT:([\s\S]+)/);
+            if (code === 0 && m)
+                resolve(m[1].trim());
+            else
+                reject(new Error(out.slice(-400) || `index exited ${code}`));
+        });
+    });
+}
+exports.syncLocalIndexTool = {
+    name: "sync_local_index",
+    description: "Analyze a LOCAL folder of footage (free on the user's GPU when local mode is on) and SYNC the tags to the " +
+        "cloud so it's searchable when planning ideas / asset-plans — WITHOUT uploading the files (tags + paths only; " +
+        "media stays local until actually used). Run this once on the user's B-roll folders so search_local_index and " +
+        "idea planning know what they already have.",
+    inputSchema: {
+        type: "object",
+        properties: { folder: { type: "string", description: "Local folder of footage to index + sync." } },
+        required: ["folder"],
+    },
+    async execute(args) {
+        if (!(0, engine_js_1.engineAvailable)("index-folder.mjs")) {
+            return "sync_local_index needs the local engine (ffmpeg + the vision model). Unavailable in this install.";
+        }
+        let r;
+        try {
+            r = JSON.parse(await runIndex(args.folder));
+        }
+        catch (e) {
+            return `Couldn't index the folder: ${e instanceof Error ? e.message : String(e)}`;
+        }
+        if (!r.items.length)
+            return `No media analyzed in ${args.folder}.`;
+        try {
+            const res = await (0, client_js_1.post)("/v1/library/local-sync", { items: r.items });
+            return (`Indexed + synced ${res.synced} local asset(s) from ${args.folder} → searchable in cloud idea-planning. ` +
+                `The files were NOT uploaded (tags only). Use search_local_index to find them by topic when planning a reel.`);
+        }
+        catch (e) {
+            return `Analyzed ${r.indexed} asset(s) but couldn't sync to the cloud: ${e instanceof Error ? e.message : String(e)}`;
+        }
+    },
+};
+exports.searchLocalIndexTool = {
+    name: "search_local_index",
+    description: "Search the user's SYNCED local footage by topic/keyword (from sync_local_index) — use when planning an idea's " +
+        "asset plan to find existing local clips that fit a script BEFORE recommending they film or AI-generate new B-roll.",
+    inputSchema: {
+        type: "object",
+        properties: { query: { type: "string", description: "Topic / keyword (e.g. 'hospital', 'sunset rooftop', 'laptop typing')." } },
+        required: ["query"],
+    },
+    async execute(args) {
+        const r = await (0, client_js_1.get)(`/v1/library/local-search?q=${encodeURIComponent(args.query)}`);
+        if (!r.items.length)
+            return `No synced local footage matched "${args.query}". (Run sync_local_index on the user's folders first, or recommend filming / AI B-roll.)`;
+        const lines = r.items.map((i) => `- ${i.path}${i.subject ? ` (${i.subject})` : ""}${i.descr ? ` — ${i.descr}` : ""}`);
+        return [`Local footage matching "${args.query}":`, ...lines].join("\n");
+    },
+};

package/dist/tools/local-mode.js

@@ -0,0 +1,84 @@
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.setLocalModeTool = void 0;
+/**
+ * set_local_mode — a persistent toggle between LOCAL analysis (free + private,
+ * a vision model on the user's GPU via Ollama) and CLOUD (Gemini). The choice is
+ * saved to ~/.viralspin/config.json, which the engine reads — so the user flips
+ * it once with no env vars to edit. (The VIRALSPIN_VLM env var still works as an
+ * explicit override.) Enabling local also verifies Ollama + the model are ready
+ * and, if not, returns the one-time setup steps.
+ */
+const fs_1 = __importDefault(require("fs"));
+const os_1 = __importDefault(require("os"));
+const path_1 = __importDefault(require("path"));
+const CONFIG = path_1.default.join(os_1.default.homedir(), ".viralspin", "config.json");
+function readCfg() {
+    try {
+        return JSON.parse(fs_1.default.readFileSync(CONFIG, "utf8")) || {};
+    }
+    catch {
+        return {};
+    }
+}
+function writeCfg(c) {
+    try {
+        fs_1.default.mkdirSync(path_1.default.dirname(CONFIG), { recursive: true });
+    }
+    catch { /* ignore */ }
+    fs_1.default.writeFileSync(CONFIG, JSON.stringify(c, null, 2));
+}
+async function ollamaStatus(model) {
+    try {
+        const r = await fetch("http://localhost:11434/api/tags");
+        if (!r.ok)
+            return { up: false, hasModel: false };
+        const d = (await r.json());
+        const base = model.split(":")[0];
+        const hasModel = (d.models || []).some((m) => m.name === model || m.name.startsWith(base));
+        return { up: true, hasModel };
+    }
+    catch {
+        return { up: false, hasModel: false };
+    }
+}
+exports.setLocalModeTool = {
+    name: "set_local_mode",
+    description: "Toggle ViralSpin video analysis between LOCAL (free + private — runs a vision model on the user's GPU via " +
+        "Ollama) and CLOUD (Gemini). The choice persists in ~/.viralspin/config.json — no env vars to edit, no restart. " +
+        "Enabling local verifies Ollama + the model and, if missing, returns the one-time setup steps.",
+    inputSchema: {
+        type: "object",
+        properties: {
+            enabled: { type: "boolean", description: "true = LOCAL (GPU/Ollama), false = CLOUD (Gemini)." },
+            model: { type: "string", description: "Optional Ollama vision model (default qwen2.5vl:7b)." },
+        },
+        required: ["enabled"],
+    },
+    async execute(args) {
+        const cfg = readCfg();
+        if (!args.enabled) {
+            cfg.vlm = "cloud";
+            writeCfg(cfg);
+            return "Switched to CLOUD analysis (Gemini). Local mode is OFF.";
+        }
+        const model = args.model || cfg.ollama_model || "qwen2.5vl:7b";
+        cfg.vlm = "ollama";
+        cfg.ollama_model = model;
+        writeCfg(cfg);
+        const { up, hasModel } = await ollamaStatus(model);
+        if (!up) {
+            return (`Local mode is ON in settings, but Ollama isn't running. One-time setup:\n` +
+                `  1) Install Ollama → https://ollama.com/download (or: winget install Ollama.Ollama / brew install ollama)\n` +
+                `  2) ollama pull ${model}   (~6GB, one-time)\n` +
+                `Once Ollama is running with the model, local analysis works automatically — no restart.`);
+        }
+        if (!hasModel) {
+            return `Local mode is ON and Ollama is running, but the model isn't pulled. Run:  ollama pull ${model}  (~6GB, one-time). Then it's ready.`;
+        }
+        return `Local mode is ON ✓ — video analysis now runs on your GPU via Ollama (${model}): free + private. Say "switch to cloud" to use Gemini instead.`;
+    },
+};

package/dist/tools/photo-picker.js

@@ -0,0 +1,31 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.startPhotoPickerTool = void 0;
+const client_js_1 = require("../client.js");
+/**
+ * Opens a Google Photos Picker session for choosing B-roll. Google does NOT allow
+ * apps to list or search a user's albums programmatically (Library API readonly was
+ * removed 2025-03-31), so the user searches/selects inside Google's own picker UI.
+ * This tool just hands back the picker link + sessionId; create_video_assembly then
+ * ingests the picks via google_picker_session.
+ */
+exports.startPhotoPickerTool = {
+    name: "start_photo_picker",
+    description: "Open a Google Photos picker to choose B-roll for create_video_assembly. Returns a link to open in your browser. " +
+        "Google does not allow listing or searching your albums programmatically — so use the picker's OWN search box to find an " +
+        "album or topic that matches your video (by person, place, event, or theme), select the photos/clips, and finish. " +
+        "Then call create_video_assembly with google_picker_session set to the returned sessionId. " +
+        "Requires a Google account connected at app.viralspin.ai/settings.",
+    inputSchema: { type: "object", properties: {} },
+    async execute() {
+        const r = await (0, client_js_1.post)("/v1/assembly/picker", {});
+        return [
+            "📸 Open this link to pick your Google Photos B-roll:",
+            r.pickerUri,
+            "",
+            "Tip: use the picker's search box to find an album/topic that matches your video — by person, place, event, or theme — then select the clips/photos and finish.",
+            "",
+            `When done, run create_video_assembly with:  google_picker_session: "${r.sessionId}"  (plus your A-roll).`,
+        ].join("\n");
+    },
+};