Human motion
intelligence
from any video.
Production-grade 3D human motion data from ordinary video, delivered by API, for the teams building games, animation, sports products and robots.
Unedited output. This reel is raw Mocai reconstruction from ordinary single camera clips. No cleanup, no keyframes, no manual retargeting.
From clip to motion, in four passes.
Phone footage works. No markers, no suits, nothing to calibrate.
Per-frame segmentation locks onto the person across the whole clip.
Full mesh and 127-joint skeleton estimated for every frame, hands included.
Occlusions repaired, body set on a solved floor so it stays put and upright.
Straight into Blender, Unreal, Maya, Unity, or your own tools.
Seventy tracked points. Every finger.
Each performer resolves to 70 tracked keypoints, pelvis to fingertips, rebuilt on every frame of your footage. The points land in hierarchy order: root first, fingers last. The mesh you saw above is draped over exactly these points, and skeleton exports carry the full 127-joint native rig: articulated spine, twist chains, and feet.
Any video in. Production-grade motion out.
Raw pose estimates jitter, tilt, float above the floor, let planted feet slide and lose limbs behind props. Mocai fuses frontier vision models and adds the motion layer that makes their output shippable: grounding, foot locking, smoothing and retargeting to your rig. Switch it off and see.
Cycling on its own. Flip any layer to take over.
Each performer stays themselves for the whole clip, even when people cross paths. You choose who to track.
A dense mesh in the subject's real proportions, rebuilt every frame. Not a stand-in wearing your motion.
Per-finger articulation comes standard, free tier included. Grips and gestures, not mittens.
A world-space trajectory on the solved floor. Walks travel, jumps land, turns stay put.
Onto Mixamo, UE5 Manny and MetaHuman, or a T-pose or A-pose first frame for retargeting by hand.
FBX, USD, BVH and GLB on a 127-joint rig, plus raw 3D keypoints as JSON or CSV.
Three levels, chosen per take. Raw keypoints are always kept untouched underneath.
Scrub the source and the 3D together on one timeline. The video is the clock.
One motion layer, every industry.
Real takes from ordinary video, read the way each industry needs them: a skinned character for engines, a humanoid for robot learning, joint angles for sport, every finger for sign language and avatars.
Studio-quality takes without a stage. The performer's own mesh, on your rig, in your engine.
Built for developers, and their agents.
Everything in the portal is an API call. Upload a clip, get a webhook when the take is ready, pull the files. Or hand the whole job to an agent.
npx @mocai/cli process clips/ uploads a whole folder, waits and downloads the results. Also on PyPI.AI agentsMCP serverClaude, ChatGPT and coding agents capture motion for you over MCP, with a sign-in instead of keys.PluginsEngine and DCC pluginsUnreal, Unity, Blender and Maya: upload clips and import takes without leaving the editor.PostmanPostman collectionEvery endpoint preconfigured with auth and examples. Make your first call in about a minute.UsageAccount APIPlan, minutes used and minutes left, for your dashboards, your plugins and your agents.Capture, by API.
One key, every request. Upload a clip, get a webhook when the motion's ready, pull the assets. No token dance. Working with AI agents? They get their own way in, below.
Postman collection, ready to runEvery endpoint preconfigured with auth and examples. Import it and make your first call in about a minute.Capture, by conversation.
Mocai speaks MCP, the open protocol AI assistants use to work with your tools. Ask Claude, ChatGPT, or your coding agent for the motion you need, and the takes come back as files.
① In Claude or ChatGPT: add api.mocai.ai/v1/mcp as a connector and sign in with your Mocai account. No keys to copy.
② On your computer: Claude Desktop, Claude Code, and Cursor can process a whole folder of clips in one step and save the exports right next to them.
③ Same account, same plan: agent takes count toward your minutes like any other, and you can disconnect an app at any time.
Track exactly who you mean.
Crowded frame? Before processing starts, Mocai shows you everyone it found. Tap the performer you want and the pipeline follows them through the whole clip. Capture one dancer from the ensemble, or all of them as separate color-coded takes.
Maya
Blender
Unity
Unreal EngineBuilt to fit your workflow.
Takes export as FBX, GLB, BVH, USD, and raw keypoints, with optional retargets onto the Mixamo, UE5 Manny, and MetaHuman rigs, so they drop straight into the tools you already use. Whether you work in Blender, Unreal Engine, Maya, or Unity, Mocai plugs right into your pipeline.
① Upload right from the editor: title, hands, performers, rigs
② Watch the take go queued → processing → ready
③ Import the take without leaving your scene. In Unity, Mixamo takes land as Humanoid clips that retarget to any avatar.
The Mocai team is incredibly responsive and truly listens to user feedback… the platform has become an essential, efficient part of our workflow. I highly recommend Mocai to anyone looking for an ultra-reliable and super-cheap animation solution.
Metered in minutes of footage.
Every plan includes the full viewer, hand capture, downloads and API access. Paid plans add more minutes, multi-person and longer clips. Building at volume? See Enterprise and API terms.
- + 1 min of footage / mo
- + Clips up to 15s
- – Single person
- + Download all formats
- + Public API + webhooks
- – Standard queue
- + 7-day retention
- + 10 min of footage / mo
- + Clips up to 30s
- + Up to 3 performers
- + Download all formats
- + Public API + webhooks
- + Priority processing
- + 30-day retention
- + 60 min of footage / mo
- + Clips up to 60s
- + Up to 10 performers
- + Download all formats
- + Public API + webhooks
- + Highest-priority processing
- + 90-day retention
Volume API access and custom terms for studios, robotics and sports teams.
- + Custom minutes, clip length and performers
- + Billed overage instead of hard stops
- + High-frame-rate processing for fast sports
- + Highest-priority processing
- + Custom retention and invoicing
Markerless motion capture, answered.
What is markerless motion capture?
Motion capture from ordinary video, with no suits, markers or multi-camera rigs. Mocai fuses frontier vision models and adds its own motion layer on top, so a single clip becomes the subject's full 3D body: a mesh and a 127-joint skeleton with fingers, ready to export.
Who is Mocai for?
Teams that need human motion as data: game and animation studios, robotics and robot-learning teams, sports and coaching products, and sign language and avatar teams. Everyone gets the same motion layer, in the outputs that fit: animation files for engines, and BVH, USD and per-frame 3D keypoints for robotics and analysis.
Do I need special hardware or a studio?
No. Any single camera works, phone footage included. Good lighting and a mostly visible subject give the best results; there's nothing to calibrate and no setup beyond pressing record.
What formats can I export?
FBX, USD, BVH, GLB, and raw 3D keypoints in meters as JSON or CSV, plus optional FBX retargeted onto the Mixamo, UE5 Manny, and MetaHuman rigs. FBX, GLB, and USD include your performer's own skinned mesh bound to the skeleton by default. Takes drop straight into Blender, Maya, Unity, or Unreal, and free plugins for all four upload clips and import takes without leaving your editor.
Does Mocai capture hands and fingers?
Yes. Detailed finger articulation is part of every take on every plan, free tier included. Grips and gestures come through, not mittens.
Can it track multiple people at once?
Yes. Everyone in frame is tracked and reconstructed as a separate, color-coded performer, and you choose exactly who to track when you upload. Multi-person capture is available on paid plans.
Can I use Mocai through an API or an AI agent?
Yes, on every plan. The REST API uploads clips and sends a signed webhook when the take is ready, the mocai CLI processes a whole folder in one command, and Mocai's MCP server lets Claude, ChatGPT and coding agents capture motion for you.
How long does processing take?
Minutes, not hours. A typical short clip is ready in a few minutes; longer clips and extra performers take proportionally longer, and you can watch the take move through the queue live. Plans are metered in minutes of uploaded video, so a 30-second clip costs 30 seconds of your monthly allowance.
Point a camera. Get the motion.
Start with a free minute of footage and have your first reconstruction moving in minutes.