Capabilities
Useful protection context with a deliberately small surface.
The production MCP exposes ten read-only tools across seven explicit scopes. It can inspect safe VaultProof operational state and explain server-side placement, but it never accepts a provider key or changes a VaultProof resource or account.
Projects and protected credentials
Review accessible projects, protection state, credential labels, providers, operational limits, and safe activity summaries without credential values or database identifiers.
Four read-only toolsUsage, activity, and alerts
Inspect bounded usage totals, minimized operational events, active signals, alert policy state, delivery state, and masked destinations.
Three read-only toolsIntegration and connection visibility
Get server-side placement guidance, check protected-route health, and review connected MCP clients and granted scopes without repository or provider access.
Three read-only toolsLeast-privilege OAuth
Connect through authorization code with PKCE S256, explicit consent, seven narrow read scopes, short-lived access tokens, and refresh-token replay detection.
Seven read scopes · no silent expansionhttps://mcp.vaultproof.dev/mcp. Tool availability still depends on an existing VaultProof account and the client or workspace policy that controls ChatGPT Developer Mode.
Connect ChatGPT
Connect an existing VaultProof account through Developer Mode.
VaultProof authorizes existing accounts through OAuth. ChatGPT opens the VaultProof sign-in and consent flow when authorization is required.
https://mcp.vaultproof.dev/mcp
-
1
Enable Developer Mode
In ChatGPT, open Settings → Security and login, then turn on Developer mode. Availability can depend on your account or workspace policy.
-
2
Add the server
Open ChatGPT Plugins, select the plus button, and create a developer-mode connection using the full URL above, including
/mcp. -
3
Review OAuth consent
Sign in to the intended VaultProof account, review the requested read-only scopes, and approve only if the client identity and permissions are correct.
-
4
Start a new chat
Add VaultProof from the tools menu, then use one of the example prompts below. If the connection metadata changes, refresh it from the Plugins page before retesting.
Tool contract
Ten read-only tools with one clear job each.
Tool schemas, scope checks, and minimized outputs keep the public surface bounded. Existing connections never silently gain a new scope.
| Tool | What it returns | OAuth scope |
|---|---|---|
vaultproof.projects.list | Accessible projects and safe operational protection status. | projects:read |
vaultproof.project.get | One project's safe configuration and protection status. | projects:read |
vaultproof.credentials.list | Protected credential labels, providers, projects, and states. | credentials:metadata:read |
vaultproof.credential.get | One protected credential's safe status, limits, and activity summary. | credentials:metadata:read |
vaultproof.usage.get | Bounded request usage without plan, price, or billing details. | usage:read |
vaultproof.activity.search | Paginated, minimized operational and audit events. | activity:read |
vaultproof.alerts.get | Signals, policy, delivery state, and masked destinations. | alerts:read |
vaultproof.integration.guide | Server-side placement guidance without repository access. | integrations:read |
vaultproof.integration.health | Protected-route configuration and health without provider responses. | integrations:read |
vaultproof.connections.list | MCP client names, granted scopes, and connection state without tokens. | connections:read |
Security boundary
Authorization is explicit, scoped, short-lived, and revocable.
The MCP endpoint authenticates the client and user before reaching authorized VaultProof metadata through a private Cloudflare service binding.
OAuth + PKCE
Authorization codes use PKCE S256, validated client metadata, issuer-bound responses, and explicit consent.
Minimized output
Raw keys, token values, internal IDs, project IDs, request IDs, and diagnostics are removed from tool responses. Safe operational timestamps appear only where a tool needs them.
Revocation controls
Refresh tokens rotate with reuse detection. Revoking a grant blocks existing access tokens and future refreshes.
Example prompts
Ask about protection state, operations, or server-side placement.
These prompts stay within the current public contract and are suitable for confirming tool selection after connection.
Show my VaultProof projects and their protection status.
List protected credential labels and providers without showing keys.
Summarize request usage and failed activity from the last 24 hours.
Show current alerts and their delivery state.
Check whether my protected route is configured and responding.
Explain where VaultProof belongs in my backend.
List the MCP clients connected to my VaultProof workspace.
Manage access
Review or revoke every connected MCP application.
Use the account connection page to inspect the approved client and scopes. Revocation immediately blocks the grant's current access tokens and future refreshes; reconnecting requires a new OAuth approval.