Primate Intelligence MCP — Video Scene Understanding for AI Agents¶
Rating: ★★ | Category: Media & Content | Transport: Streamable HTTP (OAuth 2.1 + Dynamic Client Registration + PKCE)
What It Does¶
Primate Intelligence gives AI agents the ability to watch and understand video content at scale. Register a video by URL, ask questions in plain English, and get deterministic yes/no/indeterminate answers with confidence scores and clip timestamps. Unlike AI models that hallucinate video descriptions, Primate Intelligence uses predictive world models — it builds a structured understanding of what's actually happening in each scene and answers questions against that model. The result: reliable, auditable answers about video content without watching the video yourself.
Why Business Operators Need This¶
Content monitoring teams spend hours manually checking competitor videos for claims, messaging changes, and compliance issues. Operations teams verifying user-generated content at scale can't watch every upload. Primate Intelligence lets your AI agent do this — "Does this competitor's product demo claim feature X?" or "Is there any adult content in these 200 user submissions?" — with deterministic answers and timestamps you can verify. For media companies, e-commerce operators monitoring competitor video content, brand safety teams, and compliance teams handling video submissions, this is a force multiplier. First MCP for AI video understanding with auditable, non-hallucinated results.
Quick Start¶
# Install via npm (stdio transport)
npx @primate-intelligence/mcp
# Remote endpoint (Streamable HTTP with OAuth 2.1 + PKCE)
Endpoint: https://api.primateintelligence.ai/mcp
Hermes Agent Configuration¶
{
"mcpServers": {
"primate-intelligence": {
"transport": "http",
"url": "https://api.primateintelligence.ai/mcp",
"auth": {
"type": "oauth2",
"authorization_url": "https://api.primateintelligence.ai/oauth/authorize",
"token_url": "https://api.primateintelligence.ai/oauth/token",
"client_id": "YOUR_CLIENT_ID",
"scopes": ["video:read", "video:analyze"]
}
}
}
}
Environment Variables¶
export PRIMATE_INTELLIGENCE_CLIENT_ID="pi_..."
export PRIMATE_INTELLIGENCE_CLIENT_SECRET="pi_secret_..."
Key Tools¶
Primate Intelligence exposes 10 tools with full MCP annotations:
| Tool | Description |
|---|---|
register_video |
Register a video by URL for analysis — supports YouTube, Vimeo, and direct MP4 links |
ask_question |
Ask a yes/no question about the video; returns answer + confidence score + timestamp |
ask_open_question |
Ask an open-ended question; returns most relevant scene + description |
list_videos |
List all registered videos and their analysis status |
get_video_status |
Check analysis progress for a registered video |
get_scene_at_timestamp |
Retrieve the scene understanding model at a specific timestamp |
get_all_scenes |
Retrieve all detected scenes with descriptions and timestamps |
compare_scenes |
Compare two scenes or two videos for differences |
search_across_videos |
Search for a concept or claim across all registered videos |
delete_video |
Remove a video and its analysis data |
Pricing¶
Free tier available with limited video registrations and question quota. Paid tiers for higher volume, longer videos, and priority processing. Check primateintelligence.ai for current plans.
Authentication¶
OAuth 2.1 with Dynamic Client Registration (RFC 7591) and PKCE. Register your MCP client at primateintelligence.ai/developers to obtain a client_id. The OAuth flow includes PKCE for security. Sessions are scoped and refreshable.
Alternatively, use the npm package (npx @primate-intelligence/mcp) for local stdio transport with API key authentication.
Source¶
- GitHub: github.com/Primate-Intelligence/primate-intelligence-mcp (0★, created 2026-07-26)
- Website: primateintelligence.ai
- MCP Endpoint:
https://api.primateintelligence.ai/mcp - npm:
@primate-intelligence/mcp
Verdict: ★★ — Pioneering Video Understanding for Agents¶
Primate Intelligence is the first MCP server for AI video understanding with auditable results. The predictive world model approach — deterministic answers instead of hallucinated descriptions — sets it apart from generic vision AI. For operators who need to monitor competitor video content, verify UGC at scale, or check video compliance without manual review, this is a genuinely new capability.
Strengths: Deterministic yes/no/indeterminate answers (no hallucination), confidence scores and clip timestamps for audit, cross-video search, OAuth 2.1 + PKCE security, 10 well-documented tools.
Limitations: Brand new (0 stars, created July 26, 2026), free tier limits unknown, video processing speed not documented, only yes/no questions are fully deterministic (open questions are probabilistic), requires videos to be publicly accessible by URL.
Best for: Content monitoring teams, brand safety operators, competitive intelligence analysts, compliance teams handling video submissions, media companies managing large video libraries.