Privacy Policy

1. Who we are

Agent Anything is a control plane for connecting the apps you use to the AI agents you already work with (for example Cursor, Claude, or ChatGPT). You connect an app once. You then decide, per agent, what that agent may call — Always, Ask, or Never.

The public product is at https://www.agent-anything.com. Agents talk to Agent Anything at https://mcp.agent-anything.com/mcp.

This policy is published by Agent Anything. It is governed by the laws of New South Wales, Australia.

2. Information we collect

We collect only what we need to run the product you signed up for. We do not scrape your connected apps for marketing profiles, and we do not run advertising pixels on these pages.

Information you give us

• Account details. Name, email address, and a password if you sign up with email. If you continue with Google (or another supported identity provider), we receive the identity token that provider issues — typically your email and display name.
• Waitlist details. If you joined the public waitlist, we store the email you submitted and whether we have sent a confirmation.
• Support mail. Anything you send to ben@agent-anything.com, including the request itself.

We never ask you to type the password of a connected app (Gmail, Notion, GitHub, and so on) into Agent Anything.

Information from apps you connect

When you Connect an app, that app’s normal OAuth (or API-key) flow runs. We store:

• Access and refresh tokens (or an API key you paste), so we can call the app on your behalf. Tokens are envelope-encrypted at rest in PostgreSQL. A data-encryption key wraps the blob; Azure Key Vault wraps that key in production.
• Connection metadata. Which listing you connected, health, last refresh error (a safe reason only — never the raw token), and the scopes the app granted.
• Per-agent grants and rules. Which agents may use the connection, and whether each tool is Always, Ask, or Never.

We do not store connected-app passwords. We do not keep raw tool arguments or upstream results in our audit log by default — activity records are metadata only (which agent, which app, which tool, the decision).

When an agent calls a tool, we decrypt the token briefly, attach it only to the upstream MCP request, and never forward your Agent Anything session or agent bearer token to that app.

Information collected automatically

• Essential cookies. A session cookie (aa_human_session) after you sign in, Firebase Auth cookies for that sign-in, and short-lived cookies used during OAuth / Connect redirects. These are required for the product to work.
• Technical logs. Request IDs, timestamps, coarse error codes, and similar operational data. Observability redacts tokens, passwords, API keys, and other obvious secrets. Redaction is a backstop, not a licence to log sensitive payloads.
• Directory metadata. Public listing text and icons in Explore. Azure OpenAI may rewrite public catalogue blurbs (not your private connected-app content) so search and cards stay readable.

We do not currently use advertising cookies, analytics pixels, session-replay tools, or a third-party marketing tag manager on the product site. If that changes, we will update this policy.

3. How we use your information

• Create and authenticate your account (Firebase Auth on Google Cloud for production human sign-in).
• Hold and refresh connected-app credentials so agents can call tools you have allowed.
• Route MCP tools/list and tools/call to the right upstream app, namespaced, under your grants.
• Enforce Always / Ask / Never and fail closed on Deny or Ask (Ask is not an interactive approval inbox in the current product; the call does not go upstream).
• Send transactional email through the connected Resend integration (verification, password reset, waitlist confirmation). We do not hold a Resend API key for that mail path.
• Operate Explore / Directory (public listings, Save, Connect readiness). Azure OpenAI is used only to enrich public listing text.
• Secure the service, prevent abuse, debug failures, and meet legal duties.

We do not sell personal information. We do not use connected-app content to train a general-purpose Agent Anything model.

4. Legal bases for processing

Agent Anything is offered from Australia. We handle personal information in line with the Privacy Act 1988 (Cth) and the Australian Privacy Principles (APPs).

In practice that means we collect and use information:

• To provide the service you asked for (your account, connections, grants, and agent calls).
• For security and abuse prevention, which is reasonably necessary to run an internet service that holds tokens.
• With consent where the law requires it (for example optional waitlist mail). You can withdraw waitlist consent by emailing us or using the unsubscribe path we send.
• To meet a legal obligation if one applies.

If you are in a region that uses “legitimate interests” or similar language, the corresponding interest is operating, securing, and improving the control plane you signed up for.

5. How we share information

We share personal information only as described here.

Service providers (sub-processors)

These companies help us run Agent Anything. They are not the apps you choose to Connect — those stay under their own terms.

We do not currently send product telemetry to Sentry, and we do not currently process payments with Stripe. Stripe appears in the Directory as a connected app you may choose to Connect; that is not Agent Anything taking card payments.

Redis holds ephemeral cache, locks, and short-lived OAuth state only — never the only copy of a credential.

AI agents you connect

When you authorise an agent (Cursor, Claude, ChatGPT, or another MCP client), that agent can call the tools you have granted. Those calls go through Agent Anything to the upstream app. The agent vendor then processes whatever the tool returns, under that vendor’s terms and privacy policy. You are responsible for what you grant each agent.

You can revoke an agent, ungrant an app, or set Never at any time. Revoking stops new calls from that agent; it does not rewind history the agent already saw.

Connected apps

Each app you Connect is a separate service. We send it the token it issued so it can perform the action you allowed. That app’s own terms and privacy policy apply to the data it holds.

Legal requirements

We may disclose information if we believe in good faith that we must (a court order, a lawful request, or to protect people from serious harm). We will limit what we share to what is reasonably necessary.

6. How long we keep data

• Account. While the account is open. After a verified deletion request we delete or de-identify remaining records except where Australian law requires us to keep them (for example a dispute or security investigation).
• Connected-app tokens. Until you disconnect the app. Disconnect deletes the credential blob. We do not keep a spare copy of the token in Redis.
• Grants, rules, and connection metadata. While the connection or agent exists, then as part of account deletion.
• Audit events. Metadata only (no args/results). Kept for a limited operational and security period, then dropped or aggregated.
• Waitlist. Until you ask us to remove the address or we shut the list.
• Mail logs. Only what the connected Resend path retains in the ordinary course of sending.

There is not yet an in-product “delete account” button. Email ben@agent-anything.com and we will handle deletion. You can already disconnect apps and revoke agents yourself.

7. How we protect your data

Other controls in the current product:

• TLS in transit on the public hosts.
• Envelope encryption for credentials (env1. blobs). Production wraps data-encryption keys with Azure Key Vault; local/test uses a dev key. Key Vault is not a folder of one secret per user token.
• Tokens, passwords, API keys, and raw sensitive MCP payloads must not be logged. Structured logs carry a request ID, not secrets.
• Custom MCP URLs are checked before connect (HTTPS, no loopback / private / cloud-metadata targets in production) to reduce SSRF.
• Always / Ask / Never is evaluated before an upstream tools/call. Deny and Ask fail closed.
• You can disconnect an app, revoke an agent, or ungrant access at any time.

No method of transmission or storage is perfectly secure. If we become aware of a breach that Australian law requires us to notify, we will notify you and the OAIC as required.

8. Your rights and choices

Under the Australian Privacy Act and the APPs you can:

• Access the personal information we hold about you.
• Correct it if it is wrong, incomplete, or out of date.
• Ask us to delete it, subject to any record we must keep.
• Complain to us first. If we do not resolve it, you can complain to the Office of the Australian Information Commissioner (OAIC).

In the product you can also:

• Disconnect an app (tokens are removed).
• Revoke an agent or ungrant an app.
• Change Always / Ask / Never on tools you have connected.
• Sign out (the session cookie is cleared).

To exercise access, correction, or deletion, email ben@agent-anything.com. We may need to verify that the request comes from the account holder. We will answer within a reasonable period.

If you are in the EEA/UK or another region with extra rights (portability, restriction, objection), email us. We will consider the request in good faith even though Australian law is the primary framework.

9. International transfers

Core product data — accounts, encrypted tokens, grants, directory rows, audit metadata — is stored in Microsoft Azure, Australia East.

Some processors necessarily handle data outside Australia:

• Google Firebase Authentication (sign-in).
• Resend (transactional email).
• Cloudflare (DNS resolution for our domains).
• Azure OpenAI is deployed in our Azure tenancy; Microsoft may process service diagnostics under its Azure terms.

APP 8 (cross-border disclosure) applies. We use these processors only to run the service. We do not sell or rent the data they see. If a processor is in a country without an APP-8 adequacy finding, the disclosure is for the service you requested and is made with this notice.

10. Children's privacy

Agent Anything is not directed at children. You must be 16 or older to create an account. We do not knowingly collect personal information from children under 16. If you believe we have, email ben@agent-anything.com and we will delete it.

11. Changes to this policy

We will change this policy when the product changes (for example when paid billing, analytics, or extra processors go live). The “Last updated” date at the top is the source of truth.

Material changes will be posted on this page. If we have an email for your account, we may also send a notice. Continued use after the update means the new policy applies, except where Australian law requires a different rule.

12. Contact us

For privacy and contract questions, use the address below.

This policy is governed by the laws of New South Wales, Australia. The courts of New South Wales have jurisdiction, subject to any non-excludable rights you have under Australian Consumer Law.