Frequently Asked Questions

What is a Transfer Oracle operator?

An operator is a licensed partner who runs the Transfer Oracle model auditing service for their own customers. You get your own branded portal, API keys, customer management dashboard, and AI agent integrations — all powered by our platform.

How do I apply?

Fill out the application form on the operator page. We review applications within 48 hours and look for domain expertise, customer base, and alignment with our mission.

What does the 14-day trial include?

• Full access to all operator tools — dashboard, customer management, usage analytics.
• Up to 5 customer seats with API access.
• 100 audit API calls included.
• Your own MCP endpoint for AI agent integration.
• No credit card required. Cancel anytime.

What does it cost after the trial?

Pricing depends on your tier. The API tier starts at $49/month and includes audit tools, account management, and usage reporting. Team ($199/mo) and Compliance ($499/mo) tiers add feature alignment, consensus checking, and priority support. Enterprise pricing is custom.

Can I try it before applying?

Yes — the public Growt API MCP is free for up to 2 audit calls with no API key. Install it via "uvx mcp-growt-audit" or "npx -y mcp-growt-audit" and run a demo audit from your AI agent.

What is MCP (Model Context Protocol)?

MCP is an open standard that lets AI assistants connect directly to data sources and tools. Instead of copy-pasting data into chat, your AI agent can call Transfer Oracle tools directly — audit models, check usage, manage customers — all from natural language.

Which AI agents support MCP?

• Claude Code (Anthropic) — full support via .mcp.json
• Cursor — full support via .cursor/mcp.json
• Windsurf (Codeium) — full support via mcp_config.json
• VS Code Copilot — support via .vscode/settings.json
• Any MCP-compatible agent — the protocol is open and growing.

Can I use multiple MCPs at once?

Yes. You can configure the Operator MCP (for your dashboard), the Customer Audit MCP (for testing audits yourself), and the Investor MCP (for board reporting) all in the same config file. Each gets its own server name.

Do my customers need to install anything?

No. Customers connect via a single URL — your branded MCP endpoint. They add one JSON block to their AI agent config and they're connected. No packages, no dependencies, no Docker.

What if my AI agent doesn't support MCP yet?

You can still use the REST API directly. Every MCP tool maps to a standard HTTPS endpoint. The MCP layer is a convenience — not a requirement.

Where do I find setup instructions?

See our MCP Setup Guide for step-by-step instructions covering all supported AI agents, all operating systems, and all 4 MCP endpoints.

How do my customers connect?

Share the Customer Audit MCP URL (mcp.transferoracle.ai/mcp) with your customers. They add it to their AI agent config with their API key, and they can start running model audits immediately.

How are API keys managed?

Each customer gets a unique API key when you onboard them. Keys are scoped to your operator account — customers can only access tools you've enabled. You can rotate or revoke keys from your operator dashboard.

Can I see what my customers are doing?

Yes. The Operator MCP gives you full visibility: which customers are active, how many API calls they've made, which endpoints they use, and daily usage breakdowns. Use "list_tenants" or "tenant_detail" from your AI agent.

What happens when a customer exceeds their tier?

API calls are rate-limited by tier (60/min for API, 300/min for Team, 600/min for Compliance, unlimited for Enterprise). Customers hitting limits get a clear error response. You'll see tier alerts in your operator dashboard.

Can I white-label the service?

Yes. Your customers see your branding, your tool names, and your MCP endpoint. The underlying platform is invisible to them. Tool names and descriptions are fully customisable from the operator config.

How do I give investors access?

Share the Investor MCP URL (investor-mcp.transferoracle.ai/mcp) with your investors. They connect their AI agent and can run read-only queries about platform health, growth trends, and revenue metrics.

What can investors see?

• Platform snapshot — total tenants, API calls, operators, estimated MRR.
• Growth trends — 30-day API usage trend and platform KPIs.
• Tier health — per-tier breakdown with estimated revenue.
• Revenue report — Stripe actuals and MRR/ARR estimates.
• Web traffic — sessions, users, bounce rate, traffic sources.
• Market overview — tenant and operator distribution.

Can investors modify anything?

No. The Investor MCP is strictly read-only. No write operations, no customer data modification, no configuration changes. It's an audit window, not a control panel.

Is individual customer data exposed to investors?

No. All data is aggregated. No tenant names, email addresses, domains, or individual usage patterns are shared. Investors see totals, averages, and trends — never PII.

Is the data encrypted?

Yes. All connections use HTTPS with TLS 1.2+. API keys are transmitted via headers, never in URLs. Data at rest is encrypted in our PostgreSQL database.

Where is data stored?

Data is stored on dedicated servers in Falkenstein, Germany (EU). We use Hetzner dedicated instances with full disk encryption. GDPR compliant by design.

What about GDPR?

We're GDPR compliant. We process only the feature vectors you send for auditing — no personal data is required. Our DPA (Data Processing Agreement) is available for download. Data residency is EU by default.

Can I audit who accessed what?

Yes. Every API call is logged with timestamp, endpoint, API key, and response status. Operators can view usage breakdowns by customer and by day. Audit logs are retained for 90 days.

What data does the audit process see?

The audit API receives numerical feature vectors (embeddings, sensor data, tabular features) — not raw images, text, or personal information. We never store your source data. Feature vectors are processed in memory and discarded after the response is returned.

What are the API rate limits?

• Free tier: 2 calls (no API key required).
• API tier ($49/mo): 60 requests/minute.
• Team tier ($199/mo): 300 requests/minute.
• Compliance tier ($499/mo): 600 requests/minute.
• Enterprise: unlimited (custom agreement).

What feature vector formats are supported?

The API accepts JSON arrays of numbers. MCP tools also accept .npy, .npz, and .csv file paths for convenience. Dimensions up to 4,096 are supported per sample.

How large can feature matrices be?

Up to 10,000 samples per dataset, 4,096 dimensions per sample, within a 200 MB total request budget. For larger datasets, contact us about batch ingestion.

What does "transfer oracle accuracy" mean?

The transfer oracle score measures how well a model's learned representations generalise to new data. A high score (>70%) means the model captures genuine patterns. A low score (<40%) with high validation accuracy is a red flag — the model memorised training data but won't transfer to production.

Can I use the API without MCP?

Yes. Every function available via MCP is also available as a standard REST API endpoint. MCP is a convenience layer — not a requirement. See the API documentation for endpoint details.

Is the MCP connection stateless?

Yes. The Streamable HTTP transport is fully stateless. Each tool call is independent. No sessions, no cookies, no WebSocket connections to maintain. This makes it reliable across firewalls, proxies, and containerised environments.