Pika MCP is a remote Model Context Protocol server that hands Claude authenticated access to your Pika Agent. One OAuth. One wallet. Fourteen-plus generation models orchestrated by a persistent persona that remembers your taste, voice, and visual signature across every session.
The one-paragraph answer, then the deeper picture.
Pika MCP is a remote Model Context Protocol server hosted at mcp.pika.me/api/mcp that lets any MCP-compatible AI client — Claude Desktop, Claude.ai, ChatGPT, Cursor, and others — authenticate against your Pika Agent and use its full creative stack from inside their own chat interface.
Behind the single endpoint sits a persistent persona you've already built on pika.me, defined by three identity files (identity, soul, style), and a federated stack of fourteen-plus generation models for video, image, voice, music, and transcription. The agent picks which model fits each request. You pay from one shared token wallet rather than fourteen separate provider invoices.
The MCP itself doesn't replace Claude - it Pikafies it. Your Pika Agent's personality, voice profile, and visual taste travel with you into Claude conversations, so a request like "make me a 6-second product demo" returns something on-brand instead of generic.
Pika MCP collapses the multi-tab AI creative workflow into a single endpoint that any MCP-compatible client can plug into.
A remote HTTPS endpoint Pika hosts. You don't deploy it — add it as a custom connector in Claude, authorize, and you're in.
A persistent persona you create on pika.me. Three identity files — identity, soul, and style — steer every model call in your voice.
Fourteen-plus generation models the agent routes between. You describe the outcome; the agent picks the model.
One shared token balance. Every generation debits at the model's own rate. Token packs run $7.99 to $149.99.
What you actually get when you wire Pika MCP into Claude.
Video, image, voice, music, transcripts — all from the same chat. The agent assembles them into one cohesive deliverable.
Three identity files (identity, soul, style) thread through every model call so output stays on-brand session to session.
Describe the outcome, not the implementation. The agent picks the right model — Sora for cinematic, Pika Video for speed.
Connect once. Bill from one token balance. No more juggling fourteen API keys and fourteen separate provider invoices.
Pre-built skills like /pika:podcast, /pika:explainer, and /pika:ugc-ads turn one prompt into a finished deliverable.
Works with any MCP-compatible client. Your Pika Agent shows up in Claude, ChatGPT, Cursor, and others — same persona everywhere.
The open-source pikastream-video-meeting skill lets your agent join Google Meet calls as an animated avatar with your face and voice.
Your Pika Agent can search websites, scan online libraries, browse music catalogs, and pull in research as creative context.
Generations aren't one-shot. The agent defines editing steps so every creative output can be refined and personalized further.
A single tool surface. Underneath, your agent routes to whichever generation model fits the shot — cinematic video, conversational voiceover, ambient score, or transcript.
Two steps — add the connector, then install the skills plugin. Your agent does the rest.
https://mcp.pika.me/api/mcpidentity_persona_read for identity, soul, and style.Pika-Labs/Pika-Plugins/pika:podcast, /pika:explainer, or /pika:ugc-ads.Pre-built skills that turn a single prompt into a finished, multi-modal deliverable.
Creator-persona ad variants with Gen-Z cadence and platform-specific cuts. Have your agent promote your project — or anything, really.
Cast your speakers, drop a written brief or URL, and let your Pikafied Claude assemble a video podcast with B-roll and chapter markers.
Feed it a URL, a GitHub repo, or a short brief. Get back a narrated explainer with on-screen text and smart cuts.
Real workflows that map cleanly onto Pika MCP's strengths inside Claude.
Generate Gen-Z-style creator ads in your brand voice. Spin up 5–10 platform-specific cuts for TikTok, Reels, and Shorts from a single brief.
Cast your speakers, drop a URL or written brief, and let your Pikafied Claude produce a full podcast with B-roll, chapter markers, and on-screen text.
Feed Claude a GitHub repo or product spec. Get a narrated explainer with smart cuts that you can ship to a landing page in minutes.
Style and soul files keep every image, thumbnail, and key art consistent — no more drifting palettes or off-tone illustrations.
Describe a scene. The agent picks Sora for cinematic shots, Pika Video for fast iteration, and stitches them with voiceover and score.
PikaStream 1.0 puts your animated agent on a Google Meet at 24 FPS / ~1.5s latency. They take notes, answer questions, and run tasks live.
MiniMax Music for ambient beds, ElevenLabs for narration. Generate, transcribe, and edit the audio bed without leaving Claude.
Your Pika Agent speaks languages you don't. Generate localized voiceover, subtitles, and copy for global audiences in one pass.
Pika Wallet token packs scale from light experimentation to heavy creator workflows.
How billing works: The Pika MCP connector itself costs nothing. Every generation debits your Pika Wallet at the underlying model's own rate — a Sora clip costs more than a Pika Video clip. Prompts and chats that run through Claude itself still hit your existing Claude subscription.
When the bundled wallet and persona memory pay off — and when they don't.
| Dimension | Pika MCP | Direct APIs |
|---|---|---|
| Setup | One OAuth, ~5 min | One key per provider |
| Billing | Pika Wallet tokens | Multiple invoices |
| Model choice | Agent picks for you | You pick explicitly |
| Persona memory | Built in | DIY system prompts |
| Cost structure | Wallet markup per token | Pay providers directly |
| Volume scaling | Caps at largest pack | Provider tiers, custom contracts |
Where Pika MCP wins, and where it doesn't.
identity_persona_read at the start of each new session.
The answers people actually search for before connecting.
The remote endpoint is https://mcp.pika.me/api/mcp. Add it as a custom connector in Claude (Settings → Connectors → Add custom Connector), authorize the OAuth flow, and grant tool permissions.
The connector itself is free. Generations are charged from your Pika Agent Wallet at the underlying model's own rate. Token packs run from $7.99 (800 tokens) to $149.99 (15,000 tokens).
A Sora 2 clip costs more tokens than a Pika Video clip. Prompts that run through Claude rather than Pika still bill against your Claude subscription as normal.
Pika MCP wraps fourteen-plus generation models behind one OAuth, one wallet, and one persona. Your agent picks the right model per task and threads your persona files through every call.
Direct API access means separate keys, separate billing, and writing your own orchestration code. Pika MCP is for creative workflows that live inside chat; direct APIs are for production pipelines that batch hundreds of clips per day.
Yes. Any MCP-compatible client that accepts a remote MCP URL can connect — Claude Desktop, Claude.ai, ChatGPT (via custom connectors), Cursor, and others. Claude is the primary integration target, but the protocol is client-agnostic.
The pika.art consumer app has free credits, but the MCP path bills exclusively against the Pika Wallet token packs. To test the MCP, either bring an existing Pika balance or buy the smallest 800-token pack at $7.99.
PikaStream 1.0 is Pika's real-time video-chat layer that lets your agent join a Google Meet as an animated avatar with your face and voice. It ships as a separate open-source Pika Skill (pikastream-video-meeting in the Pika-Labs/Pika-Skills repo).
It bills at roughly $0.20–$0.275 per active minute against the same Pika Wallet, but it lives in Claude Code or terminal-agent environments rather than the Claude Desktop connector list.
The Pika-Labs/Pika-Skills repo on GitHub is Apache 2.0 licensed and currently exposes one production skill: pikastream-video-meeting.
The slash workflows — /pika:podcast, /pika:explainer, /pika:ugc-ads — are managed Pika Plugins installed from the Pika marketplace, not from the open repo.
Video: Pika Video, Sora, Veo 3, Kling, MiniMax Video, Seedance 2.0, Remotion, Video Frames. Image: Gemini Image (nano banana), ChatGPT Images 2, SeedDream Image. Audio: ElevenLabs, MiniMax Music, MiniMax Voice, OpenAI Whisper. Plus a range of LLMs powering the agent's reasoning.
You do. Pika's terms grant you full IP ownership of any chats, images, videos, audio, and other content you generate — provided the inputs and likeness are yours to use and the generation aligns with their Terms of Service and Privacy Policy.
No. Pika states that personal inputs are used to create your experience on the platform only. Your likeness and inputs are not used to train other people's Pika Agents or any general-purpose models. Full details are in Pika's Privacy Policy.
Five minutes. One endpoint. Your Pika Agent — wherever you work.