
Claude Code Workflows
Engineers using Claude Code for investigation, OSINT, or content moderation can drop the MCP server into their config and let Claude run searches conversationally.
A hosted Model Context Protocol server that gives Claude, Cursor, and any MCP-compatible client five typed face-search tools. One Bearer token, zero infrastructure.
5 tools • streamable HTTP • same API keys as the REST API
Trusted by thousands worldwide
Install once in your MCP client config, then let your agent drive face searches on command.

Log in to the dashboard, create a key scoped to search:read and search:write, and copy the plaintext. The same key works for REST and MCP.

Paste the MCP server block into your client config. Restart your client. Five tools appear automatically: submit_face_search, get_search_status, list_searches, delete_search, get_credits.

Ask your agent to 'find where this photo appears online' and it handles the rest: submit, poll, present results, flag low-confidence matches, and refuse unsafe requests.
Built to the current MCP spec with guardrails that AI agents actually need.

Built to the current MCP spec with guardrails that AI agents actually need.
submit_face_search, get_search_status, list_searches, delete_search, get_credits — each with a zod input schema, clear description, and mapped errors that Claude can explain to the user.
Uses the same API keys as the REST API. One key, two surfaces. Revocation propagates to both within one request, so a lost key can be killed once from the dashboard.
When a tool call supplies imageUrl, the server resolves the hostname, rejects private CIDR ranges, forbids HTTP (HTTPS only), disables redirects, and enforces a 10-second timeout and 10 MB body cap.
The server derives the client idempotency key from each MCP call's requestId. Transport-level retries automatically dedupe, so an agent that replays a tool call gets the same jobId without double-charging credits.
MCP tool calls share the same rate-limit bucket as REST. A leaked key cannot be used to flood both surfaces independently — the ceiling is shared and enforced at Redis speed.
Instead of bare protocol errors, every failure returns CallToolResult with isError: true, a human-readable text content block (for Claude to explain to the user), and a structuredContent payload for programmatic clients.
When your AI agent needs to do face search as part of a larger workflow.

Engineers using Claude Code for investigation, OSINT, or content moderation can drop the MCP server into their config and let Claude run searches conversationally.

Cursor-style IDE agents can submit face searches as part of larger pipelines: scrape a page, extract a face, call FaceSearch, annotate the results — all in one run.

Any MCP-compatible host works. Build a voice assistant, a Discord bot, or a custom agent framework and plug FaceSearch in via the same Bearer-auth HTTP endpoint.
Feedback from teams wiring FaceSearch into their agent workflows.
"The shared Bearer auth between MCP and REST was the feature that sold me. I test with curl during development and switch to Claude Code with the same key, no re-plumbing."
"SSRF protection on imageUrl saved me from a dumb mistake on day two. I passed a redirect URL by accident and the server rejected it cleanly instead of hitting our internal metadata service."
"Structured error payloads mean Claude can actually explain rate limits to our analysts. Before, failures were opaque protocol errors; now they come back as plain English with actionable details."
Choose the plan that fits your needs. No hidden fees, cancel anytime.
For professionals and power users
Starter
$6
$2.00/credit
Basic
$24
$0.80/credit
Standard
$69
$0.69/credit
Pro
$199
$0.66/credit
Enterprise
$599
$0.60/credit
What agent builders ask before wiring the server into production.
Drop five lines into your MCP client config and your agent can run face searches conversationally.
No credit card required