The Secureframe MCP Server is now in public beta. It offers read-only, real-time access to your Secureframe compliance data via AI and developer tools that support the Model Context Protocol (MCP). 

No more hunting through dashboards—just ask natural-language questions and get instant answers on compliance status, controls, vendors, devices, and more.

Read more about MCP in our Blog! 

Key Features

• 11 read-only endpoints, including:

• Lucene-style filtering, so you can query by framework, owner, test status, risk level, etc.
• Secure and scoped:
  • Read-only access only, safeguarding production data.
  • Does not permit writes or destructive actions.

Common Use Cases

1. Check failing SOC 2 controls
   “Show me failed endpoint‑security controls in SOC 2.” AI returns current failing items and insights on owners and remediation timelines
2. Vendor risk analysis
   “Which third‑party vendors are high‑risk?” AI surfaces flagged vendors and risk status for quick triage
3. Audit prep
   “What are our most recent failed ISO 27001 tests?” Instantly fetch test results to streamline audit readiness.
4. Access reviews
   “List inactive users with system access.” Spot stale accounts and support access clean-up workflows 

How It Works

1. Host: The MCP server must be self-hosted in your environment.
2. Authorize: Grant the MCP Server to interact with AI clients using Model Context Protocol.
3. Query: Using tools like Claude, Cursor IDE, or ChatGPT with MCP support, you can run natural language prompts such as “list failing controls” or “show high-risk vendors.”
4. Respond: AI fetches live data from the mapped endpoints and presents it conversationally.

Technical Details

• Built on the open MCP standard.
• Offers 11 scoped, read-only endpoints covering controls, tests, devices, users, vendors, frameworks, integrations, and repository mappings
• Supports powerful filters using Lucene query syntax for precision.
• Read-only architecture ensures full safety—no modifications to production data.

Getting Started

1. The newly built MCP server feature is now available as a public beta — the code is open to all customers on GitHub and must be self-hosted.
2. Connect your MCP server to your preferred AI client.
3. Start querying your compliance data conversationally.
4. For setup help or troubleshooting, reach out to Secureframe support via Support Portal or email

Best Practices

• Always verify AI output before acting—the data is live, but AI can misinterpret.
• Use scoped filters in prompts (e.g. “SOC 2 only”) to target your requests.
• Treat this as read-only access—no data or configurations can be changed via MCP.

Why It Matters

• Faster insight: Get real-time compliance updates without digging into dashboards.
• AI‑powered workflows: Use your tools—IDE, ChatGPT, Slack—to interact with compliance data naturally.
• Enhanced visibility: Know exactly where you stand without context-switching or manual effort.

Next Steps

• Explore the MCP spec and sample queries.
• Try example prompts like:
  • “List all failed SOC 2 endpoint controls.”
  • “Show third-party vendors flagged as high‑risk.”
  • “Give me the most recent ISO 27001 test failures.”
• Stay tuned for future enhancements like write‑only tasks or deeper integrations.

Need help enabling or using the MCP server? Visit the Secureframe Support Portal or email support@secureframe.com

Platform Compatibility & Hosting Requirements

Some customers use AI-enhanced development environments such as dust.tt, Claude Code, or editors like Zed and Cursor. 

If you're working in these environments, there are a few key considerations:

• The Secureframe MCP server must be self-hosted, similar to how other providers operate.

• Shared agent platforms (e.g., dust.tt) may not support the ability to host local servers, which can prevent successful integration.

• Known integration attempts:

  • Following the README setup for editors like Zed/Cursor may not work, as these editors may not detect your Python virtual environment.

  • Using pipx to make the server callable from a single command hasn't consistently resolved the issue.

  • Running main.py with API keys directly from an activated venv may not result in a working server, especially if the editor is unaware of the environment.

If you are using a shared or cloud-based agent (like dust.tt), and cannot host the MCP server yourself, integration may not be currently feasible.

GitHub repo: https://github.com/secureframe/secureframe-mcp-server

Frequently Asked Questions (FAQ)

How do I set up the MCP server if I’m using a platform like dust.tt or Zed?

If you're using platforms like dust.tt, Claude Code, or editors such as Zed or Cursor, please note that the MCP server must be self-hosted in order to function properly. This setup is similar to how other platforms like Vanta manage their MCP integrations.

Customers using dust.tt and shared agents have reported difficulty integrating directly due to editor limitations in recognizing Python virtual environments. Attempted workarounds such as using pipx for packaging or running main.py directly from an activated venv have not yielded consistent results.

Because shared agent environments (like those on dust.tt) may not support persistent or custom local servers, you will likely need to host the Secureframe MCP server yourself to enable integration. If self-hosting is not an option, this may pose a limitation for using the MCP server in such environments.

For setup, you can find the GitHub repository here: https://github.com/secureframe/secureframe-mcp-server

We’re continuing to evaluate platform compatibility and will update this page with any supported alternatives in the future.