Pika MCP
Complete Guide

For the last two years, making AI video meant living in browser tabs. You wrote a prompt in one tool, generated an image in another, cloned a voice somewhere else, then stitched it all together by hand. Every model had its own login, its own credits, and its own quirks. Pika MCP collapses that sprawl into a single conversation. It is an official connector that brings Pika's creative engine inside an AI assistant you already talk to - Claude, Cursor, Codex, OpenClaw and other MCP-compatible agents - so you can describe what you want and watch it get made, without leaving the chat.

Image credit: www.pika.me/mcp

What Pika MCP is

MCP stands for Model Context Protocol - an open standard, originally introduced by Anthropic, that gives AI assistants a common way to talk to outside tools and data. People often describe it as a "USB-C port for AI": instead of every product building a bespoke integration, a tool exposes its capabilities through one standard interface, and any MCP-compatible client can plug in. Pika MCP is Pika's implementation of that standard.

Concretely, it is a hosted server living at mcp.pika.me/api/mcp. When you add it to a client like Claude, your assistant gains a set of Pika "tools" it can call on your behalf - generate a video, edit an image, write a score, clone a voice, cut a clip. Pika describes this as "Pikafying" your agent: turning a general assistant into a creative partner that can actually produce finished media rather than just describe it.

There is a second layer that makes Pika's version distinct. Pika MCP is tied to your Pika Agent - a persistent creative identity with its own personality, voice, look, and memory. When you connect the server, the agent's identity travels with it, so the assistant doesn't just gain raw generation power; it gains a consistent point of view and a memory of what you've made before.

Think of it as the difference between renting a film crew for a day and hiring a creative director who already knows your brand. The tools are the same; the continuity is what changes.

Pika MCP Preview the Skills

Pika MCP gives your AI agent creative video and content-making skills for launch videos, app store assets, podcasts, trend formats, founder storytelling, UGC ads, and social media campaigns. These skills help creators, startups, and brands turn simple ideas into polished visual content using ready-made creative commands.

1. Explainer

Create clear explainer videos that break down a product, service, idea, or feature in a simple and engaging way. Command:/explainer

Image credit: www.pika.me/mcp

2. Build-A-Brand

Generate brand storytelling content, startup visuals, identity concepts, and creative launch materials for founders and businesses. Command:/build-a-brand

Image credit: www.pika.me/mcp

3. Baseball Trend

Create videos inspired by viral sports-style scoreboard and baseball broadcast trends for social media content. Command:/baseball-trend

Image credit: www.pika.me/mcp

4. App Store Screens

Design app store preview visuals and screenshots that showcase app features in a clean, promotional format. Command:/app-store-screens

Image credit: www.pika.me/mcp

5. App Sizzle

Create high-energy app promo videos that highlight key features, benefits, and user experience in a stylish launch format. Command:/app-sizzle

Image credit: www.pika.me/mcp

6. UGC Ads

Generate user-generated-content-style ads that feel natural, personal, and perfect for TikTok, Reels, Shorts, and paid campaigns. Command:/ugc-ads

Image credit: www.pika.me/mcp

7. Podcast

Create podcast-style scenes, interview visuals, and branded podcast content for creators, hosts, and storytelling formats. Command:/podcast

Image credit: www.pika.me/mcp

8. Kiss Cam

Generate fun kiss-cam-style trend videos with a playful social media look, ideal for viral entertainment content. Command:/kiss-cam

Image credit: www.pika.me/mcp

9. Founder Product Video

Create founder-led product videos that explain a product, share the story behind it, and build trust with the audience. Command:/founder-product-video

Image credit: www.pika.me/mcp

Why use it

The core argument is consolidation. A modern creative project touches many models - one for video, one for images, one for voice, one for music. Pika MCP aggregates them behind a single agent, and lets you either name a specific model or let the agent choose the right one for the task. That alone removes a lot of friction.

But the bigger shift is interface. Pika calls its design "AI-native - not a GUI wrapped in an API." The tools are built to be driven by an agent, not clicked through by a human. So instead of filling out a form with resolution, seed, and duration fields, you say "make me a fifteen-second product demo from this app store link," and the agent assembles the steps. Here is who benefits most:

• Founders and indie makers who need launch videos, app store screenshots, and a brand kit without hiring an agency or learning five tools.
• Marketers producing UGC-style ads, explainers, and social cuts at volume, on demand, inside the assistant they already write copy in.
• Developers who want to call generation steps from Claude Code or Codex while they build, turning a GitHub repo or brief into an explainer in the same window.
• Creators who want a consistent on-brand voice and look across everything they publish, thanks to the persistent agent identity.

How to set it up

Setup has two parts: connect the server, then install the companion skills. The flow below is for Claude, which is the most common client, but the same connector works across MCP-compatible apps.

Step 1 - Connect the MCP server

1. Open Claude.ai → Settings → Customize → Connectors.
2. Click the + to add a new connector.
3. Search for Pika to find the custom connector (or point it at the server URL https://mcp.pika.me/api/mcp).
4. Authorize the connection. This runs an OAuth-style sign-in that links your Pika account without ever pasting a raw key into the chat.
5. Grant tool permissions according to your comfort level - you decide which Pika tools the agent may call.
6. Ask the assistant to pull in your Pika identity so it adopts your agent's persona, voice, and style.

Step 2 - Install the Pika Skills plugin

Once the server is connected, ask your agent to install the companion skills bundle. In a coding client like Claude Code or Codex this is a one-line command:

npx skills add Pika-Labs/Pika-Plugins

That installs ready-made Skills - packaged workflows like Explainer, Podcast, and UGC Ads - that turn a single sentence into a finished, multi-step output. To get newly released Skills later, you re-run the same command to update. One useful detail: you can't add the custom connector from the Claude iOS app directly, but if you connect it on Claude.ai it syncs to mobile automatically, credentials and all, so you don't authenticate twice.

Features & toolkit

The reason to connect Pika rather than a single-model server is breadth. Behind the one connector sits a large, growing toolkit and a library of underlying models, all callable in natural language.

The creative toolkit

On the video side: text-to-video, image-to-video, video-to-video, reference video, extend, re-cut, modify, and lip-sync. On the image side: text-to-image, reference image, image edit, upscale, inpaint, background removal, face swap, and style extraction. On audio: music generation, text-to-speech, voice cloning, and transcription. And for finishing: auto-captions, burned subtitles, trimming, stitching, overlays, transitions, and prompt remixing. In practice you rarely name these directly - you describe an outcome and the agent sequences the right tools.

The model library

Pika MCP routes across many of the best generation models rather than locking you to one. Reported models include Pika Video, Kling, MiniMax, Veo 3, Sora, Seedance, and Remotion for video; Gemini Image (nano banana), ChatGPT Images, and SeedDream for stills; and ElevenLabs, MiniMax, and OpenAI Whisper for voice, music, and transcription. You can pin a specific model or let the agent pick based on the job.

Ready-made Skills

Skills are the highest-leverage layer - pre-built recipes that chain many tools into one result. The launch set and growing catalog include an Explainer (turn a URL, repo, or brief into a narrated walkthrough), a Podcast (cast characters and hand them a topic), UGC Ads (fast, authentic-feeling promos), a Founder Product Video (script, talking-head avatar, cloned voice, lip-sync, captions, publish), App Sizzle and App Store Screens, Build-a-Brand, and playful trend formats. New Skills roll out regularly and update with the same install command.

How to prompt it

Because the toolkit is agent-driven, prompting Pika MCP feels less like operating software and more like briefing a collaborator. A few habits get better results:

• Describe the outcome, not the settings. "A 15-second vertical UGC ad for my running-shoe brand, upbeat, with captions" beats a list of parameters. The agent fills in the steps.
• Give it source material. A URL, a GitHub repo, product screenshots, or a written brief anchors the result. Skills like Explainer and Founder Product Video are built to ingest these.
• Name a model only when you care. If you specifically want Veo 3 or Kling for a shot, say so; otherwise let the agent route to the best fit.
• Iterate line by line. Treat the first output as a draft. "Make the second scene brighter and re-cut to 9:16" refines without starting over.
• Lean on the identity. Once your Pika Agent's persona is loaded, ask it to keep things "in my brand voice" and it carries tone across generations.

The mental model that works best: you're the director giving notes, and the agent is the crew. You set intent; it handles the machinery.

Pricing & credits

Pika MCP splits billing across two meters, and understanding the split avoids surprises. Anything that actually generates or edits media through Pika draws from your Pika Agent Wallet - measured in Credits, visible in your Pika wallet. Meanwhile, the ordinary chatting, reasoning, and prompting that runs through your host agent (Claude, Cursor, Codex) bills against that product's own subscription. In short: the assistant's "thinking" is on your existing plan; the creative "rendering" is on Pika.

Pika Agent credits are bought as token packs. At the time of writing the published tiers are:

• 800 tokens - $7.99
• 2,000 tokens - $19.99
• 4,000 tokens - $39.99
• 8,000 tokens - $79.99
• 15,000 tokens - $149.99

New accounts typically start with a batch of tokens to try things out. Note that balances from older Pika video accounts don't carry into the Pika Agent experience - it's a separate product, so you fund a fresh wallet. Because pricing and starting allowances change, treat these figures as a snapshot and confirm current numbers in your Pika wallet before committing to a big project.

API for developers

Beyond the chat connector, Pika exposes a developer API so you can bring Pika Skills and Powers into your own product or agent - including frameworks like OpenClaw - without setting up a full Pika Agent first. You register at pika.me/dev/login, generate an API key, and skip the consumer onboarding. Select Skills are also published openly on GitHub under Pika-Labs/Pika-Skills, which is handy if you want to read how a Skill is structured or adapt one.

For builders, this is the difference between using Pika and embedding it. The MCP connector is the fast path for everyday creation inside an assistant; the API is the path for wiring Pika's generation into an app, a pipeline, or a bot you ship to other people.

Security & privacy

Handing an AI agent the ability to generate media on your account raises reasonable questions, and the Pika MCP design answers several of them by construction.

• OAuth authorization, not pasted keys. Connecting runs a proper authorize step, so you're not dropping a raw secret into a chat window where it could be logged.
• Granular tool permissions. During setup you choose which Pika tools the agent may call, so you can keep a tight scope rather than granting everything at once.
• Real-time moderation. Pika scans inputs and outputs for sensitive or high-risk content, blocks flagged material, and offers reporting on every interaction. The broader app experience is age-gated at 17+.
• Your data isn't training fuel. Pika states it doesn't use your likeness or inputs to train other people's agents or general-purpose models - your inputs power your experience only.
• You own the output. The IP for what you generate is yours, provided you have the rights to the inputs you used.
• Private by choice. An agent can be set to private so only people you approve can see or use it.

Two cautions still apply. First, like any MCP connector, only grant permissions you understand, and be deliberate about which client you authorize. Second, the same tools that power legitimate work - voice cloning, face swap, talking-head avatars - can be misused, so only ever generate likenesses you have permission to use, and be honest when content is AI-made.

Advantages

Pulling the threads together, here's where Pika MCP genuinely stands out:

• One context for everything. Generate, edit, and finish media without leaving your assistant - no tab-juggling, no manual handoffs between tools.
• Many models, one door. Access a broad model library and let the agent route, or pin a model when it matters.
• Persistent identity and memory. Your Pika Agent's personality, voice, and history carry across sessions, which keeps output on-brand.
• Skills do the heavy lifting. Pre-built recipes turn a sentence into a finished explainer, ad, or product video.
• Cross-client and cross-device. Works in Claude, Cursor, Codex, and OpenClaw, and syncs between web and mobile once connected.
• Developer-friendly. An API and open Skills repo mean you can embed Pika, not just use it.

Limitations

It's a young product built on fast-moving models, so it's worth being clear-eyed about the rough edges:

• Two bills to track. The split between Pika Credits and your host-agent subscription is logical but easy to forget; heavy generation can spend tokens quickly.
• It's still experimental. Pika itself frames the agent experience as early-stage - first attempts may not nail your intent, and good results often take iteration.
• Generation timeouts. Video tasks are slow by nature, and some MCP clients impose time limits that can cut long renders short. This is a known friction across video MCP servers, not unique to Pika.
• Separate account needed. The Pika Agent experience doesn't inherit your old Pika video account or its balance.
• Mobile setup quirk. You connect on the web first; the iOS app can't add the custom connector itself (though it syncs afterward).
• Quality varies by model and task. Routing across many models means results aren't perfectly uniform; some jobs shine, others need a different model or a re-prompt.

Pika MCP vs other MCP tools

Pika isn't the only creative MCP server. Tools like the PiAPI server expose models such as Midjourney, Flux, Kling, and Luma to MCP clients; Pictory's server targets script-and-template business video; and MiniMax offers its own MCP for TTS, image, and video. What separates Pika is the combination of a broad, multi-vendor model library, an agent identity with persistent memory, and ready-made Skills that ship as complete workflows rather than raw model calls. Where a single-vendor server gives you one model's output, Pika gives you a directed result assembled from several. The fuller side-by-side is in the comparison table further down this page.

Example workflows

To make this concrete, here are a few end-to-end jobs people run through the connector, and how each one tends to play out in a single conversation.

Launch-day product video

You paste your app store link or a short brief and ask for a founder-style product video. The Founder Product Video Skill writes a script from your material, generates a talking-head avatar, clones a voice, lip-syncs the delivery, adds captions, and hands back a publish-ready cut. You review, ask for a tighter intro, and re-render just that section.

Explainer from a repo

A developer points the Explainer Skill at a GitHub repository and asks for a 60-second walkthrough. The agent reads the source material, drafts a narration, generates supporting visuals, and assembles a captioned segment - useful for a README header video or a demo on a landing page.

A week of social ads

A marketer asks for five UGC-style ads for one product, each with a different hook, all vertical and captioned. The agent batches them, keeps the brand voice consistent via the loaded identity, and the marketer cherry-picks the two strongest to post. Because the identity persists, next week's batch picks up where this one left off.

The bottom line

Pika MCP is best understood as a bet on a particular future: one where you don't operate creative tools, you brief an agent that operates them for you. For founders and marketers who need a steady stream of on-brand video, image, and audio without assembling a stack, it removes an enormous amount of busywork. For developers, the API turns Pika into something you can build on. The trade-offs - the dual billing, the experimental edges, the video timeouts - are real but mostly the growing pains of a new category.

If you already live inside Claude, Cursor, or Codex, connecting Pika MCP costs a few minutes and changes what "ask and receive" means in that window. You stop describing the video you wish you had and start getting it. That shift - from talking about media to making it in the same breath - is the whole point.