# SightRadar face recognition MCP server

> Face recognition MCP server for Claude, ChatGPT, Cursor and VS Code. Index photos, find a person from one selfie, compare and detect faces. OAuth or API key.

- Connector URL (streamable HTTP): https://mcp.sightradar.com/mcp
- Sign-in: OAuth 2.1 through sightradar.com (email and password, Google or an email link), or `Authorization: Bearer frs_...` with a SightRadar API key
- Local server: `npx -y sightradar-mcp` (npm package sightradar-mcp, MIT, Node.js 20+)
- Prices: no MCP fee; normal API prices from prepaid credits, 100,000 credits = US$1; index, search, register selfie and detect cost 93 credits; search by selfie id and each batch photo cost 62; compare costs 93 per image
- Setup guide: https://sightradar.com/docs/mcp

## Connect

- Claude (claude.ai, Claude Desktop): Settings, Connectors, Add custom connector, paste https://mcp.sightradar.com/mcp
- ChatGPT: Settings, Apps & Connectors, Create, paste the URL, choose OAuth
- Claude Code: `claude mcp add --transport http sightradar https://mcp.sightradar.com/mcp`, then `/mcp` and Authenticate
- Cursor (~/.cursor/mcp.json) and VS Code (.vscode/mcp.json): add a server with url https://mcp.sightradar.com/mcp

## Tools (39)

### Read (Read collections, batches, webhooks, balance and usage (free))

- `list_collections`: List collections
- `get_collection`: Get a collection
- `get_deletion_status`: Check a collection deletion
- `list_batches`: List batch jobs
- `get_batch`: Get batch status
- `get_batch_photos`: Get per-photo batch results
- `list_webhooks`: List webhooks
- `list_webhook_failures`: List failed webhook deliveries
- `get_balance`: Get credit balance
- `get_usage`: Get usage

### Search (Search, detect and compare faces (spends credits, stores nothing))

- `search_faces`: Search a collection by face
- `search_by_point_id`: Search by a stored face
- `detect_faces`: Detect faces
- `compare_faces`: Compare two faces

### Write (Create collections, index photos, register selfies, submit batches, add webhooks (spends credits))

- `create_collection`: Create a collection
- `recreate_collection`: Recreate a purged collection
- `index_photo`: Index a photo
- `register_selfie`: Register a selfie
- `submit_batch`: Submit a batch job
- `requeue_failed_batch_photos`: Retry failed batch photos
- `create_webhook`: Add a webhook
- `replay_webhook_failures`: Replay failed webhook deliveries

### Delete (Delete collections, photos and webhooks)

- `delete_collection`: Delete a collection
- `restore_collection`: Restore a deleted collection
- `delete_photo`: Delete a photo's faces
- `restore_photo`: Restore a deleted photo
- `delete_webhook`: Disable a webhook

### Account (Manage API keys, profile, notifications, auto-recharge, orders and team)

- `list_api_keys`: List API keys
- `create_api_key`: Create an API key
- `revoke_api_key`: Revoke an API key
- `get_profile`: Get account profile
- `update_profile`: Update account profile
- `get_billing_settings`: Get billing settings
- `update_auto_recharge`: Configure auto-recharge
- `get_invoice`: Download an invoice
- `get_notification_settings`: Get notification settings
- `update_notification_settings`: Update notification settings
- `manage_notification_recipient`: Manage alert recipients
- `list_team_members`: List team members

## Questions

**What is a face recognition MCP server?** An MCP (Model Context Protocol) server gives an AI assistant tools it can call. This one connects the assistant to the SightRadar face recognition API, so it can index photos, search faces by selfie, compare two faces and detect faces when you ask in plain language.

**Which AI apps does it work with?** Any MCP client that supports remote servers: Claude (claude.ai, Claude Desktop and Claude Code), ChatGPT, Cursor, VS Code, Windsurf and others. Clients that only run local servers can use the open-source sightradar-mcp package from npm.

**Does the AI app see my API key?** No. When you approve a connection, sightradar.com creates a key for it and hands it to this server directly. The app only receives OAuth tokens that work against mcp.sightradar.com, and you can disconnect it at any time.

**What does it cost?** There is no MCP fee. Each tool call is a normal API call at the normal price, paid from prepaid credits (100,000 credits = US$1). Indexing a photo, searching by selfie, registering a selfie or detecting faces costs 93 credits; a repeat search by selfie id and each batch photo cost 62. New accounts start with free credits.

**Can I limit what the assistant can do?** Yes. Choose permission groups when you approve. A Read-only connection cannot spend credits or change data, and tools outside the groups you allow are not offered to the assistant at all.

**Can it read photos from my computer?** The hosted server takes image URLs or base64 data. The local sightradar-mcp server can also read image files from folders you allow.