Pika Labs Ships MCP to Bring Its Full Creative Engine Into Claude

Pika Labs just shipped two things at once: a fun new visual effect and a serious developer integration. The Voxel-It Skill converts any uploaded photo into a blocky, Minecraft-style scene, and the Pika MCP lets you pipe Pika's entire creative engine directly into Claude, Codex, Cursor, or any other MCP-compatible agent. One is a viral content trick; the other is a full agentic media pipeline.

Block party: what Voxel-It actually does

The Voxel-It Skill is part of Pika's growing library of one-tap photo effects, sitting alongside existing tools like Pikaffects (melt, squish, cake-ify). Pika's platform lets you turn any photo into a reality-bending video with just one tap. Voxel-It specifically converts the background, the foreground object, or both into a voxel aesthetic , the chunky, cube-based visual style made famous by Minecraft , while keeping any people in the frame looking fully human. That last detail matters: the skill is smart enough to apply the style transformation selectively, so you don't end up with a blocky, unrecognizable face where a person used to be.

The practical use cases are fairly obvious:

• Social content and memes with a gaming aesthetic
• Turning product photos or landscapes into stylized game-world visuals
• Creating branded assets with a retro-pixel or voxel look
• Generating thumbnails or promotional art for gaming-adjacent content

Pika's effect skills are fairly rigid in what they do , you upload a photo and the effect is applied, with presets rather than open-ended prompting. Voxel-It follows that same pattern: it's a targeted, preset-driven transformation, not a general-purpose style editor. If you want fine-grained control over which blocks appear where or the exact palette used, you'll hit a wall quickly.

The bigger story: Pika as an MCP server

The more technically significant announcement is the Pika MCP. MCP (Model Context Protocol) is an open standard originally introduced by Anthropic that gives AI assistants a standardized way to call external tools , think of it as a universal plugin interface for agents. Pika MCP collapses the usual multi-tool sprawl into a single conversation, bringing Pika's creative engine inside an AI assistant you already use , Claude, Cursor, Codex, OpenClaw , so you can describe what you want and watch it get made without leaving the chat.

Before, you had to switch between web tools and coding, but by connecting your Pika Agent to your existing MCP-compatible AI account, you can now do it all in one context , type a prompt into your Claude interface and have all Pika's tools at your fingertips.

The tool surface exposed through the MCP is broad. Pika MCP can help create and edit images, videos, music, audio files, voiceovers, and more. The full toolkit includes:

• Text-to-video, image-to-video, video-to-video
• Lip sync, face swap, inpainting, background removal
• Image upscale, style extraction, reference image generation
• Music generation, text-to-speech, voice cloning
• Auto captions, subtitle burning, video trimming and stitching

The Pika MCP applies your Pika Agent's qualities , like personality and persistent memory , to compatible AI accounts, and connects to Pika's ever-growing library of models, letting you select a specific model or letting the agent automatically choose based on context. The model roster behind the scenes is wide: it includes Seedance 2.0, GPT Images 2.0, and more. The full list also covers Kling Video, MiniMax Video, Veo 3, Sora, ElevenLabs, and OpenAI Whisper , essentially an aggregator that routes your request to whichever model fits best.

How to connect it

Setup is two steps. First, add the MCP server endpoint to your agent. For Claude, that means going to Settings → Customize → Connectors and searching for Pika, or pointing it at https://mcp.pika.me/api/mcp. There's nothing to self-host , you add it as a custom connector in your MCP client and authorize it. Then install the companion skills plugin:

npx skills add Pika-Labs/Pika-Plugins

The official Pika Claude Code plugin ships 3 curated slash commands , /pika:podcast, /pika:explainer, and /pika:ugc-ads , backed by 42 atomic media tools. Skills also activate from natural language , saying "make a podcast about https://pika.art" fires /pika:podcast automatically. Supported clients include Claude (web and Claude Code), Cursor, Codex, Hermes, and OpenClaw.

On pricing: generations draw Pika Credits from your wallet, while your assistant's chat usage bills to your existing Claude, Cursor, or Codex plan. New accounts get a batch of trial tokens to start. For the Pika Agent app itself, token packs range from $7.99 for 800 tokens up to $149.99 for 15,000.

What this means for builders

The interesting design choice here is that the Pika MCP is AI-native design , not a GUI wrapped in an API. All available Pika tools are designed specifically for AI agents. That's a meaningful distinction. Most creative tools bolt on an API after the fact; Pika is building the agent interface as the primary surface. The result is that you can chain complex multi-step media pipelines , scrape a URL, write a script, generate a talking-head video, add captions, export , all within a single Claude Code session, without touching a browser.

The open-source Pika-Plugins repo on GitHub is worth bookmarking. Any skill-aware agent , Claude Code, Cursor, Codex, OpenCode, Cline, plus 50+ more auto-detected by vercel-labs/skills , can pick up the plugin. And because the skills are just SKILL.md files with Python scripts, they're self-contained modules that tell the AI agent when to activate, how to use them, and include step-by-step instructions the agent reads and follows. That makes them forkable and extensible for anyone who wants to build on top of Pika's API.

Voxel-It is a fun addition to Pika's growing effects library, but the MCP integration is the part worth watching. As more creative models get routed through a single agent interface, the friction of multi-tool media production keeps dropping , and Pika is positioning itself as the creative middleware layer that sits between your coding agent and the actual generation models.