Log in

MCP server

Connect your AI assistant to your Trust Center

Move your trust page over from another platform, keep it current, and read what your visitors are doing, from Claude or any MCP client. One tenant per connection, your own role as the ceiling, and nothing publishes without a person.

https://mcp.delivertrust.io/mcp

Move your trust page over from another platform

Already on SafeBase, Drata, Conveyor, UpGuard or Vanta? Point your assistant at the public page and it builds the DeliverTrust version: branding, certifications, sub-processors, controls, FAQ and custom pages. You review the plan before anything is written, and you publish.

  1. Connect your AI assistant and choose the tenant you are moving into.
  2. Give it the public URL of your current trust page and ask it to migrate.
  3. Review the plan it returns: every write, in order, plus the gaps it found.
  4. Let it apply the plan, then publish yourself from the admin app.

Try: “Migrate our trust page from https://trust.example.com into DeliverTrust”

  • SafeBase
  • Drata
  • Conveyor
  • UpGuard
  • Vanta

Document files stay behind each platform’s NDA and never migrate; the document inventory does. Only public content is read. Both are stated again in the limits below.

Get started

No API keys. You sign in with your DeliverTrust account and choose what the connection may do.

Claude (web and desktop)

  1. Open Settings, then Connectors.
  2. Choose Add custom connector.
  3. Name it DeliverTrust and paste the server URL.
  4. Sign in with your DeliverTrust account when the browser opens, pick the tenant, review the permissions and choose Authorize.
https://mcp.delivertrust.io/mcp

Claude Code

  1. Add the server from the terminal, then run /mcp inside a session to sign in.
claude mcp add --transport http delivertrust https://mcp.delivertrust.io/mcp

Other MCP clients

  1. Any client that supports remote MCP servers over HTTP with OAuth can connect. Register the server URL in its configuration; the sign-in and consent steps are the same.
{
  "mcpServers": {
    "delivertrust": {
      "url": "https://mcp.delivertrust.io/mcp"
    }
  }
}

The tools 24

7 read tools and 17 write tools, grouped exactly as the consent screen groups the permissions behind them. Your assistant only sees the ones its connection was granted. This list is generated from the server, so it is never behind it.

Your connection

  • Connection contextread
    get_auth_context

    Report which DeliverTrust tenant this connection is authorized for, who authorized it, and what it is permitted to do.

  • Authorizable tenantsread
    list_tenants

    List the DeliverTrust tenants this person could authorize a connection for, and which one this connection is bound to.

Read your trust pages

  • Read trust pageread
    get_trust_page

    Read the tenant's whole trust page as its editors see it: branding, layout, the controls, sub-processor and certification libraries, document metadata (never file contents), custom pages and the access-request workflow's content.

  • Platform catalogsread
    get_catalogs

    DeliverTrust's curated catalogs of controls, certifications and sub-processors, for picking well-known entries to add to this tenant's page with the set_* library tools.

  • Plan a migrationread
    migrate_trust_page

    Plan a migration of a public trust page on SafeBase/Drata, Conveyor, UpGuard or Vanta into this tenant: reads the source, mirrors its certifications, sub-processors, controls, FAQ, overview and custom pages as the ARGUMENTS of the write tools (set_branding, set_certifications, set_subprocessors, set_controls, the page and logo tools, set_layout), and reports every gap - gated content, documents (inventory only), in-progress claims, dead links.

Read access requests

  • Access requestsread
    get_access_requests

    The tenant's document access requests: the queue awaiting a human decision (who asked, for which documents, when) and the recent history.

Read visitor analytics

  • Trust page analyticsread
    get_analytics

    How the trust page is doing over a trailing window (7d, 30d, 60d or 90d): page views, unique visitors, top referrers, visitor actions, and the access-request funnel.

Edit branding (logo, colors, contact email)

  • Set brandingwrite
    set_branding

    Set the tenant's wordmark name, colours and header band.

  • Set logo from URLwrite
    set_branding_logo

    Set the tenant's hero/wordmark logo from an image URL (PNG, JPEG, SVG or WEBP, 1 MB max).

  • Set favicon from URLwrite
    set_favicon

    Set the public page's browser-tab favicon from an image URL (PNG, JPEG, SVG or a real .ico, 1 MB max; an .ico is converted to PNG).

Edit trust page content

  • Replace controlswrite
    set_controls

    Replace the tenant's controls library (categories of named controls with descriptions) AND put every category and item on the Controls Summary section of the page.

  • Replace sub-processorswrite
    set_subprocessors

    Replace the tenant's sub-processor register: every vendor, with the source's own values (name, what it processes, category, location, website).

  • Replace certificationswrite
    set_certifications

    Replace the tenant's certifications and compliance marks (SOC 2, ISO 27001, ...), written as the tenant's own with the source's names.

  • Replace layoutwrite
    set_layout

    Replace the trust page's layout: every section's position, size, visibility and content (hero title/subtitle/contactEmail, overview heading/body).

  • Create pagewrite
    create_page

    Create a custom page on the trust page: a rich-text page (Terms, Privacy, a security overview) or an FAQ page (question groups).

  • Update pagewrite
    update_page

    Replace a custom page's editable fields: send the WHOLE page as read from get_trust_page with your edits (page.id required; page_type cannot change).

  • Attach vendor logowrite
    set_subprocessor_logo

    Attach a vendor's own logo (PNG, JPEG, SVG or WEBP URL, 1 MB max) to one entry of the sub-processor register by vendor_id.

  • Attach badgewrite
    set_certification_logo

    Attach a badge image (PNG, JPEG, SVG or WEBP URL, 1 MB max) to one certification by certification_id, so the mark renders the badge rather than a monogram.

Remove pages

  • Delete pagewrite
    delete_page

    Delete a custom page by id.

Edit request workflow and notification channels

  • Set request workflowwrite
    set_request_workflow

    Set the access-request workflow's content: blocked email domains (send the full list), the generic-mailbox block, the NDA text, the three visitor emails and the visitor-facing copy (each sent whole).

Manage documents (upload, edit, remove)

  • Upload documentwrite
    upload_document

    Upload a document (SOC 2 report, pentest summary, policy) from a URL (pdf, docx, xlsx, pptx, png, jpg, md, txt; 25 MB max) and list it under a category on the page's Documents section.

  • Update documentwrite
    update_document

    Change a document's name, description or access gates (nda_mechanism, approval_mechanism) by document_id.

  • Delete documentwrite
    delete_document

    Permanently delete a document by document_id and remove it from the Documents section.

Manage access requests (approve, reject, revoke)

  • Decide access requestwrite
    approve_access_request

    Approve or reject a visitor's pending document access request (decision = approve | reject), for every pending document or the doc_ids given.

How authorization works

OAuth 2.1 with PKCE, the standard the major AI clients speak. Tokens last an hour under a connection that lasts 30 days of use and can be revoked at any time.

  1. Sign in

    Your AI client opens a browser window to app.delivertrust.io. You sign in exactly as you do for the admin app, MFA included. The client never sees your password.

  2. Pick one tenant

    A connection is bound to a single Trust Center, chosen on the consent screen. Every tool call is checked against that choice, so an assistant cannot wander into another tenant you belong to. To work on a second tenant, connect again and pick it.

  3. Choose the permissions

    Read access is always granted. Writing is opt-in by area: content, branding, documents, page removal, the request workflow, access requests. The permissions you grant are a ceiling under your own role: an Editor cannot grant an assistant more than an Editor can do.

  4. Work, and keep the last word

    The tools are described so that an assistant reads before it writes and confirms the change with you first. Every write lands in your Audit Logs with the same actor and detail as a hand edit, and the connection itself is logged when it is made, refreshed and revoked.

What it does not do

Stated plainly, because a limit you find out about mid-migration is worse than one you read here first.

  • It cannot publish your trust page

    Publishing and unpublishing are not grantable over MCP, and neither are changing your URL or custom domain, managing members, or deleting a tenant. An assistant can prepare a page; a person publishes it in the app.

  • Document files never migrate

    Every platform keeps its files behind an NDA or an access request. When you bring a trust page over, the document inventory comes with it and the files do not. You upload those yourself.

  • Only public content migrates

    The migrate tool reads what a visitor can read. A FAQ that a platform lists but does not answer publicly, or a sub-processor detail behind a login, is reported as a gap rather than invented.

  • A vendor security score is not carried over

    UpGuard shows a security score that has no equivalent in DeliverTrust. The migrate tool reports it as a deliberate omission and never writes a number in its place.

  • A Vanta FAQ page beyond the overview is reported as a gap

    Vanta pages are read through a real browser. The overview, certifications, sub-processors and controls are read; a FAQ page holding more than the overview is reported with counts, not partially imported.

  • Served badge images expire after 30 minutes

    Compliance badges rasterized from a Vanta page are served at short-lived URLs. Apply a migration plan while it is fresh. Past the expiry the image fetch is refused, and running the migration again produces a fresh plan with fresh images.

  • Confirmation before a write is best-effort

    The tool descriptions ask the assistant to confirm each change with you before writing. DeliverTrust cannot force a client to do so, which is why the permissions you grant, and your own role, are the hard limits.

  • You revoke your own connections

    Connections belong to the person who made them. An Administrator cannot revoke another member's connection; that member does it from Connected apps, or their access ends when their membership does.

Disconnecting

Every connection you have made is listed under Settings › Connected apps in the admin app, with the client, the tenant and when it was last used. Revoke ends every token under it immediately. By authorizing a connection you confirm you are authorized to act for that company and accept the Terms of Service.

Log in to the admin app

Questions

Is my trust page content sent to the AI provider?

When you ask your assistant something, it calls a DeliverTrust tool and the result goes back to the assistant to answer you. What the AI provider does with that is governed by your agreement with them, not with DeliverTrust. Grant read-only access if that is all you need.

Can one connection cover several tenants?

No. A connection is bound to one tenant at consent, on purpose. An assistant that could choose the tenant per call is one that can name the wrong one confidently. Connect once per tenant you want to work on.

What does a migration actually do?

It reads the public trust page you point it at (SafeBase or Drata, Conveyor, UpGuard, Vanta), maps it onto a DeliverTrust trust page, and returns a plan: the exact writes, in order, plus a summary and a list of gaps. Nothing is written until you have reviewed the plan and the assistant applies it with the write tools you granted. Publishing stays with you.

How do I disconnect?

Open Settings › Connected apps in the admin app and revoke the connection. Every token under it stops working immediately. Removing the connector inside your AI client is not guaranteed to revoke it on our side; Connected apps is the place that is.