Primateintelligence MCP server
{
"mcpServers": {
"primateintelligence": {
"command": "npx",
"args": [
"-y",
"@primate-intelligence/mcp@latest"
],
"env": {
"PRIMATE_API_KEY": "<primate_api_key>",
"PRIMATE_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"primateintelligence": {
"command": "npx",
"args": [
"-y",
"@primate-intelligence/mcp@latest"
],
"env": {
"PRIMATE_API_KEY": "<primate_api_key>",
"PRIMATE_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"primateintelligence": {
"command": "npx",
"args": [
"-y",
"@primate-intelligence/mcp@latest"
],
"env": {
"PRIMATE_API_KEY": "<primate_api_key>",
"PRIMATE_BASE_URL": "https://..."
}
}
}
}{
"servers": {
"primateintelligence": {
"command": "npx",
"args": [
"-y",
"@primate-intelligence/mcp@latest"
],
"env": {
"PRIMATE_API_KEY": "${input:primate_api_key}",
"PRIMATE_BASE_URL": "${input:primate_base_url}"
}
}
},
"inputs": [
{
"id": "primate_api_key",
"type": "promptString",
"description": "Primate api key",
"password": true
},
{
"id": "primate_base_url",
"type": "promptString",
"description": "Primate base url",
"password": true
}
]
}{
"mcpServers": {
"primateintelligence": {
"command": "npx",
"args": [
"-y",
"@primate-intelligence/mcp@latest"
],
"env": {
"PRIMATE_API_KEY": "<primate_api_key>",
"PRIMATE_BASE_URL": "https://..."
}
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y @primate-intelligence/mcpversion 0.4.0
What it does
MCP server for the Primate Vision API that gives AI agents video scene understanding capabilities. Agents can register videos, ask questions in natural language, and receive deterministic yes/no/indeterminate answers with confidence scores and clip timestamps. Connects via remote OAuth endpoint or local stdio with API key.
- Ask a question about video content (e.g., 'Is there a person in this video?') and get a deterministic answer with confidence score and timestamp references.
- Validate a question against a video before analysis to check if it's assessable and see the estimated cost, without consuming credits.
- Batch multiple questions on the same video with per-analysis cost reduction, then poll for results until completion.
Tools this server exposes
| Tool | Does |
|---|---|
| create_video_from_url | Register a video for analysis from a public HTTPS URL. |
| create_analysis | Ask a question about a registered video and receive an answer with confidence and clip timestamps. |
| validate_analysis | Dry-run a prompt to check assessability and preview estimated cost without consuming credits. |
| create_analysis_batch | Submit 2–10 questions on one video at once, with reduced cost for analyses after the first. |
| get_analysis | Fetch the current status and result of an analysis. |
| wait_for_analysis | Poll an analysis until it reaches a terminal state, returning the result and retry information. |
| list_models | List available analysis models. |
| get_usage | Retrieve credit balance and usage meters for the current period. |
| get_credits | Retrieve credit balance and per-analysis transaction ledger. |
| get_test_fixture | Get a stable test fixture for verifying integration works without consuming quota. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.