Skip to main content

Secureframe MCP Server: AI-Driven Compliance Insights

Written by Brady Price

Connect AI tools to Secureframe

The Secureframe MCP Server connects AI assistants and developer tools to your Secureframe workspace using the Model Context Protocol (MCP). Clients like Claude Code, Claude Desktop, and Cursor can query and take action on your compliance data in natural language.

Every request runs as the user tied to the connection and respects that user's existing Secureframe permissions.

Note: The server exposes 100+ tools across 35+ categories (controls, tests, vendors, users, devices, risks, SSP, and more). Browse the full catalog in the Secureframe MCP Server docs.


Key Features

  • Natural language access to compliance data (for example, "show me failing SOC 2 controls")

  • Full read and write access for create, update, delete, and archive actions available via the API

  • Same tool surface as the Secureframe REST API (100+ tools across 35+ categories)

  • Permission-aware requests that mirror the connected user's access in Secureframe

  • Hosted infrastructure, so there is no need to host it yourself

  • Region-specific endpoints for US and UK workspaces

  • Sign in with your Secureframe account using OAuth, so there are no credentials to generate or manage. Or use an API key for headless setups.


Getting Started

  1. Choose the MCP URL for your Secureframe data region (US or UK).

  2. Connect your preferred MCP client and sign in with your Secureframe account, or use an API key if your setup cannot open a browser.

  3. Restart the client, then start querying and taking action on your compliance data.

The sections below walk through each step in more detail.

Connect to your region

Use the URL that matches your Secureframe data region. The examples in this article use the US URL. Substitute the UK URL if that is where your workspace lives.

Region

MCP URL

United States

United Kingdom


Authentication

There are two ways to authenticate. Both run every request as your Secureframe user, with your existing permissions.

Sign in with Secureframe (recommended)

Point your MCP client at the URL for your region with no credentials at all. The first time it connects, it opens your browser so you can sign in to Secureframe and approve access, and the client receives its own token. There is nothing to generate, copy, or store yourself, and no secret ends up in a config file.

This is standard OAuth 2.1, so most MCP clients, including Claude Code, Claude Desktop, and Cursor, handle it automatically.

What happens when you sign in:

  1. Sign in to Secureframe (email and password, SSO, or a social login, including any MFA step). Skipped if you are already logged in.

  2. Choose a company, if you belong to more than one company that can connect MCP clients. With a single eligible company, this step is skipped.

  3. Approve access. The consent screen shows the client requesting access, the account you are signed in as, and the company the connection will act in.

Your browser then returns to the client, which stores its own token. You will not be asked to sign in again unless access is revoked.

The client is only ever granted your own access. It can see and change exactly what you can in Secureframe, in the company you selected, nothing more.

Who can connect a client: your membership in the company must be active, and at least one of the following must be true: you are an admin or super admin, your role allows you to create API keys, or you already own an active API key.

Sessions: an access token is valid for one hour and renews silently in the background, so you are not asked to sign in every hour. The renewal credential rotates each time it is used and stays valid until access is revoked.

API key

For setups with no browser, such as CI jobs, scripts, or servers, authenticate with a Secureframe REST API key and secret instead. Generate a key in Company Settings > API Keys, then send the key and secret in an Authorization header, separated by a space:

Authorization: <API_KEY> <API_SECRET>

Prefer environment variables rather than pasting credentials into a shared config file:

export SECUREFRAME_API_KEY="your_api_key"

export SECUREFRAME_API_SECRET="your_api_secret"

Important: Treat your API secret like a password. Anyone with it can act as your user in Secureframe.


Connecting a Client

Point your MCP client at the Secureframe hosted endpoint for your region. MCP servers load when the client starts, so restart the client after you add the server.

Claude Code

Add the server with a single command. No credentials needed:

claude mcp add --transport http secureframe https://mcp.secureframe.com/

Then run /mcp inside Claude Code, select secureframe, and choose Authenticate. Your browser opens to sign in to Secureframe and approve access, and the connection completes on its own.

Note:--transport http is required. Without it, the CLI treats the argument as a local command to launch. --scope is optional: the default (local) applies to the current project only, while --scope user makes the server available across all your projects.

To connect with an API key instead, when no browser is available, add the Authorization header using the environment variables from above:

claude mcp add --transport http secureframe https://mcp.secureframe.com/ --header "Authorization: ${SECUREFRAME_API_KEY} ${SECUREFRAME_API_SECRET}"

A client configured with an Authorization header skips the sign-in flow and uses the key for every request.

Other MCP clients

For clients configured with JSON (Claude Desktop, Cursor, and similar), point an HTTP MCP server at the endpoint:

{ "mcpServers": { "secureframe": { "type": "http", "url": "https://mcp.secureframe.com/" } } }

The client will prompt you to sign in the first time it connects.

To use an API key instead, add the same Authorization header:

{ "mcpServers": { "secureframe": { "type": "http", "url": "https://mcp.secureframe.com/", "headers": { "Authorization": "<YOUR_API_KEY> <YOUR_API_SECRET>" } } } }


Available tools

These tools wrap the Secureframe REST API. Query filters and parameters available on an API endpoint are also available on the matching tool. Connected clients receive each tool's parameter schema automatically.

Here is a taste of the categories and tools available today:

  • Controls and tests: list_controls, get_control, list_tests, update_test, create_test_evidence

  • Vendors and third-party risk: list_vendors, archive_vendor, list_third_party_risk_management_vendors

  • Users and access: list_users, update_user, list_user_accounts, link_user_account

  • Devices: list_devices, get_device

  • Risks and POA&M: list_risks, list_poam_items, create_poam_item, update_poam_item

  • Frameworks and SSP: list_frameworks, plus SSP report, section, policy, role, and vendor tools

  • Comments and evidence: create_comment, list_comments, get_evidence

  • Trust Center: list_trust_center_requests, update_trust_center_request

Note: Risks and tests are available through MCP today. Company Policies module content is not exposed through MCP tools yet. SSP policy tools in the catalog are for SSP and FedRAMP workflows, not your general company policy library.

In total there are 100+ tools across 35+ categories. Browse the full catalog, including read vs write labels for each tool, in the Secureframe MCP Server docs.


Common Use Cases

  • Check failing controls: "Show me failed endpoint security controls in SOC 2."

  • Vendor risk analysis: "Which third-party vendors are high risk?"

  • Audit prep: "What are our most recent failed ISO 27001 tests?"

  • Access reviews: "List inactive users with system access."

  • Device and agent health: "Which devices are missing the Secureframe Agent?"

  • Risk and POA&M follow-up: "List open high risk items that still need owners."

  • Take action in Secureframe: "Assign myself as owner of this test," or "Add a comment to this control requesting evidence."


Best Practices

  • Review what the assistant is about to do before confirming, especially for updates, deletes, or archive actions.

  • Prefer signing in with your Secureframe account over an API key when your client supports it, since there is no secret to manage or rotate.

  • If using an API key, scope it to a user with the permissions needed for the tasks you want the assistant to perform, and prefer environment variables over shared config files.

  • Treat AI output as a starting point. Verify important findings and actions before relying on them.


Frequently Asked Questions (FAQ)

Do I need to host the Secureframe MCP server myself?

  • No. Secureframe provides a hosted MCP server. Point your client at the regional URL (US or UK) and sign in with your Secureframe account, or authenticate with an API key and secret. You do not need to run or deploy the server yourself for standard Claude, Cursor, or similar clients.

Which authentication methods are supported today?

  • Two: signing in with your Secureframe account using OAuth (recommended, no credentials to manage), and an API key plus secret in the Authorization header (for headless setups like CI or scripts). Most modern MCP clients, including Claude Code, Claude Desktop, and Cursor, support the sign-in flow automatically.

Does Secureframe MCP work with Dust.tt?

  • It depends on Dust's connector requirements. Secureframe MCP supports OAuth sign-in and API key authentication.

  • If Dust can complete the Secureframe OAuth flow, or can send an Authorization header with your API key and secret, try connecting to the hosted MCP URL for your region.

  • If Dust cannot connect that way, use the Secureframe REST API with an API key instead. See the Secureframe Developer Portal.

Can I read company policies through MCP?

  • Not yet. Company Policies module content is not exposed through MCP tools today. You can query risks and tests through MCP. SSP policy tools in the catalog are for SSP and FedRAMP workflows, not your general company policy library. Browse the current tool list in the Secureframe MCP Server docs.

Can I query the risk register and tests through MCP?

  • Yes. Risk tools such as list_risks and get_risk, and test tools such as list_tests, get_test, and update_test, are available through the hosted MCP server. Access still follows the permissions of the connected user.

What if my AI platform cannot connect to MCP yet?

  • Use the Secureframe REST API with an API key from Company Settings > API Keys. Many agent platforms can call HTTP APIs even when they cannot complete an MCP OAuth flow. Point the agent at the API documentation and provide a key with the least access needed for the workflow.

Need help? Contact Secureframe support via the Support Portal or email [email protected].

Did this answer your question?