AskPortco Connector Data Policy
Draft for review. Not legal advice. Plain-English summary for operating partners and their IT and compliance teams. Checked against the built system in the C-04 security review (2026-10-22); to be reviewed by a lawyer before it's published. Bracketed items are placeholders.
Effective date: [EFFECTIVE DATE] · Operated by [LEGAL ENTITY NAME] · Contact: [CONTACT EMAIL]
The short version
- Read-only. The connector can look things up. It can't change, delete or send anything.
- Your firm only. Each firm's access token sees only that firm's portfolio.
- Free tier = public data only. No access to any portfolio company's systems is needed or used.
- Questions are logged (which tool was used and what was asked, not the answer text), to keep the service secure and understand what you're interested in. We say so up front.
- No AI on our side for the connector. Your own assistant (Claude, ChatGPT) writes the answers; the connector only returns your firm's scores and findings.
- You can revoke access at any time, yourself, with your report link.
What the connector can see
| Tier | Data the connector reads | Access needed from you |
|---|---|---|
| Free | Scores and findings built from public information: website performance data, public DNS and email-security settings, detected website technology, TLS and security headers, public web archive history, public job postings | None |
| Hub: Google Analytics 4 (paid, per company) | Aggregated website reports: traffic, sources, engagement, conversions. No ability to change settings. | A read-only (Viewer) Google sign-in, granted per portfolio company |
| Hub: Google Search Console (paid, per company) | Search performance (queries, clicks, impressions) and indexing status. No ability to change settings. | A read-only Google sign-in, granted per portfolio company |
| Custom connectors (financials, CRM, code, project tools) | Scoped case by case, and can be deployed in your own environment | Agreed in a statement of work |
How access works
- The connector runs at mcp.askportco.com. Your firm connects through OAuth 2.1 (with PKCE): on first connect, a consent page asks for your firm's private report link, which identifies your firm. We store only a hash of the connector credential, never the credential itself.
- Every request is checked against your firm's token. A query can't return another firm's data, and this is tested before launch.
- Hub connections to Google use read-only permissions, which you can see and revoke in your Google account at any time.
- Rate limits apply per firm and per IP address.
What we log, and why
Retention periods are targets; automated deletion is being put in place before launch.
| Logged | Why | Kept for |
|---|---|---|
| Your questions: the tool used and what you asked (connector); for web chat and email Q&A, once offered, the question and our answer | To improve answers and understand what matters to your firm so we can follow up | Target: 24 months, then deleted or anonymized |
| Technical logs: time, firm, tool called, answer length, errors | Security, troubleshooting, enforcing limits | Target: 90 days |
| Hub data pulled from Google | To answer your questions and produce monthly scorecards | While the Hub subscription is active, then deleted within 30 days of cancellation |
AI models
The connector calls no AI model on our side. When you use it inside your own assistant (Claude, ChatGPT), that assistant writes the answer from the scores and findings the connector returns, and the assistant's own data terms apply to the conversation.
Web chat and email Q&A (when offered) use an AI model provider ([confirm at launch]: Anthropic and/or OpenAI) to draft answers from your question and your firm's data. Email answers are reviewed by a person before they're sent.
We don't use your data to train AI models, and we use provider terms under which it isn't used for their training either.
Where data lives
Data is stored with Supabase (database) and processed on Cloudflare (hosting and the connector). Data regions: [CONFIRM REGION, e.g. US]. Data is encrypted in transit; encryption at rest is provided by these platforms.
Subprocessors
Cloudflare, Supabase, Google, PostHog, Resend, Cal.com, Stripe, Instantly, and Anthropic and/or OpenAI. The Privacy Policy says what each one does. We will update this list before adding a new subprocessor that handles customer data.
Revoking access
- Revoke it yourself from your report page using your report link, or email [CONTACT EMAIL] and we'll do it. Access stops within minutes: the grant is deleted at once, and any already-issued access token expires within an hour.
- Remove the AskPortco connector from your AI assistant's settings.
- For Hub connections, also remove AskPortco's access in the Google account's security settings. Your Hub data is then deleted within 30 days.
Reporting a security issue
Email [CONTACT EMAIL] with the subject line "Security." We'll acknowledge it within 2 business days. Please don't publicly disclose the issue until we've had a reasonable chance to fix it.
Questions your IT team will ask
Does AskPortco need access to our portfolio companies' systems? Not for the free tier, which uses public data only. Hub connections are read-only, per company, and granted by someone with authority at that company.
Can the connector change anything? No. Every tool it exposes is read-only.
Can other firms see our data or questions? No. Access is scoped to your firm's token, and this is tested before launch. Benchmarks compare anonymized aggregates only. No firm or company names are shared.
Is our data used to train AI? Not by us. We also use AI provider terms that exclude your data from their training.
What gets stored if we use the connector in Claude or ChatGPT? On our side, which tool was called, what was asked, the answer length and technical logs (see the logging table above). Not the answer text. The AI assistant may also keep the conversation under its own terms.
Can this run in our own environment? For custom connectors to sensitive data such as financials or code, yes. That's scoped in a statement of work.
Do you have SOC 2? Not yet. This document describes our current controls. [Update if this changes.]
How do we shut it off? See "Revoking access" above. Revocation takes effect within minutes.