Skip to main content

Release Notes – August 2026

Explore two product updates: connect AI tools to permitted Trust Center content through MCP, and use extra questionnaire guidance to improve AIQA answers and reduce manual review.

Written by Jane Baik

Table of Contents

Trust Center account review via MCP is now in early access

The Problem

Procurement and security teams are already using AI agents to speed up vendor Trust Center reviews. Without a standard, sanctioned interface, reviewers have had to build their own scrapers or ask vendors for one-off integrations.

A typical review still requires manually clicking through the portal, browsing products and sections, searching for a specific certification or report, and downloading the document before an AI tool can read it.

The Solution

Trust Center visitors can now connect their AI tool directly to a vendor’s Trust Center through MCP (Model Context Protocol). There is no separate credential to create. Authorization reuses the same email-based access flow visitors already use to view the Trust Center. Once connected, the AI tool can use read-only tools to search, browse, and retrieve documents.

Access is scoped exactly like the web portal. The AI tool cannot see anything the visitor could not already access there.

Notes & Callouts

  • This feature is designed for external reviewers who are evaluating a vendor’s Trust Center.

  • The vendor you connect to is determined by the Trust Center you’re signed in to when you approve the connection in your AI tool—not by the URL you enter.

  • If you have access to multiple vendors, you must sign in and authorize access separately for each vendor.

  • All four tools are read-only.

    • tc_access lists every Trust Center the visitor can access.

    • tc_explore navigates a Trust Center’s structure, products, pages, and content groups.

    • tc_search finds specific facts across documents, Knowledge Base entries, content items, and published updates.

    • tc_document_download returns a time-limited, signed download link for a document.

  • Supported clients include Claude, ChatGPT, Cursor, Claude Code, Codex CLI, Windsurf, and VS Code.

  • All vendors use the same MCP URL: https://app.safebase.io/api/trust-center/mcp.

  • The feature is rolling out gradually to public Trust Centers, so it might not be available in every Trust Center immediately.

Learn more at SafeBase Vendor Trust Center MCP User Guide (Early Preview)

AIQA now supports a new column in questionnaires

The Problem

Some security questionnaires include an additional column next to the question or answer cell. This column provides instructions for the person completing the questionnaire, such as how to format an answer or which information to include.

Because we previously had no way to ingest this column, we dropped that context and our AI could not use it when answering questions. In some cases, answers were formatted incorrectly or did not include the right information, which created additional manual review and correction work.

The Solution

Customers can now select an Additional Instructions column when mapping a questionnaire. When selected, AIQA uses the column to help answer each question and takes the extra details into account.

This results in higher-quality answers and less manual work for people reviewing and completing security questionnaires. We also display the additional instructions next to the question in the question cell, with a clear visual distinction.

Notes & Callouts

  • AIQA automatically selects a context column when its header matches one of these names: context, instructions, additional notes, or additional instructions.

  • The feature currently works with Excel and CSV-formatted spreadsheets.

Related Articles
Did this answer your question?