The LogicGate MCP Server is currently in development and not available to all customers. We expect to release this feature to our customers in our second release of September 2026.
How to connect your organization's AI tools to Risk Cloud data using the Model Context Protocol (MCP) including what's supported, how permissions are enforced, and what your AI governance team needs to review.
- What is the LogicGate MCP Server?
- What You Can Do With the LogicGate MCP Server
- What's Supported in the LogicGate MCP Server
- Requirements
- How to Connect
- Supported AI Tools
- How Authentication Works
- How Permissions Are Enforced
- Frequently Asked Questions
- Troubleshooting
- Information for AI Governance Teams
What is the LogicGate MCP Server?
The Model Context Protocol (MCP) is an open standard for connecting AI tools to the systems that hold your data. Think of it as USB-C for AI: instead of building a custom integration for every AI tool you use, Risk Cloud exposes its capabilities once in a standard shape, and any MCP-compatible AI tool can use them.
The Risk Cloud MCP Server is a hosted endpoint in your own Risk Cloud environment. When you connect an AI tool to it, that tool can look up and read your Risk Cloud data, such as applications, workflows, fields, and records, all scoped to your individual permissions, and then answer questions or summarize findings in plain language.
A distinction that matters for governance: the Risk Cloud MCP Server is a data source, not an AI feature. It does not contain a large language model and it does not send your data to a model provider. The AI model is whichever tool you connect running under your organization's own AI agreements and controls, for example Claude or Copilot. This is different from LogicGate AI and GRC Agents, which process data through LogicGate's model providers. See Information for AI Governance Teams for more details.
What You Can Do With the LogicGate MCP Server
Once connected, you can ask your AI tool questions that require live Risk Cloud data. Examples drawn from customer conversations:
- "What third-party assessments are waiting on me?"
- "Show me all my pending issues."
- "What vendors are coming up for renewal?"
- "What are my most critical risks?"
- "What risks don't have assessments yet?"
- "What records are overdue, and who is assigned to them?"
- "Summarize the results from our maturity assessments into an executive summary."
- "Review these records and summarize the key findings."
- "Verify this intake form was filled out correctly."
Because MCP is a standard, your AI tool can combine Risk Cloud data with data from other systems you've connected in a single conversation whether it's your contract management system, a ticketing tool, or an internal knowledge base. That cross-system composability is one of the most common reasons customers ask for MCP.
What's Supported in the LogicGate MCP Server
This first release is read-only.
| Supported | Reading applications, workflows, steps, fields, field values, and records. Counting and aggregating records. Searching to resolve names to records, workflows, and applications. |
| Not supported | Creating records, updating field values, or moving records through workflow steps. Creating or modifying applications, workflows, or steps. Bulk record import or export. User management, account settings, or any admin/configuration operation. |
| Planned for a later release | Write access to records (create, update field values, submission). Referencing linked records. Ability to reference dashboards, visual reports, and saved table reports. |
Read-only access is enforced in three independent ways, not just by the choice of tools we published:
- A curated tool list. Only the read tools listed below are registered with the server.
- A runtime safeguard. Every tool call is checked against a read-only verb allow-list before it executes. A tool that isn't a read operation is rejected even if it were registered by mistake.
- Per-tool authorization. Each tool requires a specific read scope on your access token, checked at call time.
Requirements
- API access. The MCP Server is included with Risk Cloud API access at no additional cost. It is not a separate paid add-on. Your user must have the MCP Access module entitlement through an assigned Role.
- An MCP-compatible AI tool. See Supported AI Tools and Setup Instructions below.
- HTTPS connectivity from the tool to your Risk Cloud environment. Some organizations route AI tool traffic through a proxy or secure web gateway; you may need to allow your Risk Cloud subdomain.
How to Connect
Your MCP server address is your Risk Cloud environment URL with /mcp appended:
https://YOUR_SUBDOMAIN.logicgate.com/mcp
For example, the address for https://grc.logicgate.com/ is https://grc.logicgate.com/mcp.
This is the address that you will use when configuring the MCP server in your AI tool. Navigating directly to this address will not allow you to connect to the MCP server. For more on connecting your tool of choice to Risk Cloud, see the Supported AI Tools and Setup Instructions section below.
Supported AI Tools and Setup Instructions
MCP is an open standard, so in principle any MCP-compatible tool can connect. In practice, support varies by tool and by version, particularly in how each one handles the OAuth login step.
| Tool | MCP Setup Instructions |
|---|---|
| Claude (desktop app) | https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp |
| Claude Code (CLI) | https://code.claude.com/docs/en/mcp-quickstart |
| Microsoft Copilot Studio | https://learn.microsoft.com/en-us/power-pages/configure/mcp-connect-clients |
| Cursor | https://cursor.com/docs/mcp |
What if my tool isn’t listed? That does not mean the tool won't work. It means LogicGate hasn't tested that specific tool end to end, so we can't promise it. Each vendor implements the OAuth callback differently, and that behavior changes between versions, so we confirm tools individually rather than inferring from the vendor's name. If you get a tool working that isn't marked verified, tell your account team because we'd like to add it to this list. If it doesn't work, tell us too, and we'll investigate.
How Authentication Works
The MCP Server uses OAuth 2.1 with PKCE, the current standard for authorizing third-party applications, rather than a shared API key.
- You log in as yourself. Authorization runs through your existing Risk Cloud login, including SAML SSO. There is no separate MCP password and no shared service account.
- Your AI tool registers itself. MCP clients register automatically using Dynamic Client Registration, so there's no manual setup for LogicGate or your admins. Each connected tool gets its own credentials, which can be revoked independently.
- Scoped access. Tokens carry narrow read-only scopes (mcp:records:read, mcp:applications:read, mcp:workflows:read, mcp:fields:read), which are then intersected with your actual Risk Cloud entitlements.
Clients that support custom HTTP headers can alternatively authenticate with a standard Risk Cloud API token as a Bearer token. OAuth is the recommended path for end users, because it ties every session to a named individual and avoids distributing long-lived tokens to third-party tools.
How Permissions Are Enforced
Every MCP session acts as you, with your permissions. This is the most important thing to understand about the MCP Server, and it's the answer to most security questions about it.
- Entitlement scoping. All reads go through the same permission-checked paths as the Risk Cloud API. What your AI tool can see through MCP is identical to what you can see in the Risk Cloud interface.
- No shared service account. Sessions are tied to a specific authenticated user, never a pooled or elevated identity unless you configure a service account for an agent you create outside of Risk Cloud.
- No client-supplied identity. The user is derived from the authenticated token. An AI tool cannot ask to act as a different user, and cannot pass an account or user identifier to widen its access.
- Session isolation. Each MCP session operates in an isolated context. Data from one session is not accessible to another session, user, or tenant.
- Single-tenant by construction. Each Risk Cloud environment runs its own instance. There is no shared multi-tenant MCP endpoint, so cross-customer data access is not possible.
- Admin control. An administrator can enable or disable the MCP Server for the environment at any time.
Frequently Asked Questions
Is the MCP Server an extra cost? No. The MCP Server is included with Risk Cloud API access at no additional cost. It is not a paid add-on.
How is this different from LogicGate AI or GRC Agents? LogicGate AI and GRC Agents are AI features LogicGate builds, using LogicGate's model providers, inside Risk Cloud. The MCP Server allows your users to use their tools and models to connect to Risk Cloud. No LogicGate-managed model is involved.
My AI tool asks for an OAuth client ID and secret. Where do I get those? You don't need them for the tools documented here, Risk Cloud registers MCP clients automatically. If your tool has no automatic-registration option and requires credentials to be entered manually, it can't connect yet. Tell your account team which tool it is.
Can an AI tool change my data through MCP? No. This release is read-only, enforced in three independent ways. Write access is planned for a later release and will be documented separately.
Will it see data I'm not supposed to see? No. Every read is scoped to your Risk Cloud entitlements. If you can't see a record in Risk Cloud, your AI tool can't see it through MCP.
Can I connect Risk Cloud data alongside data from other systems? Yes, that's one of the main reasons to use an MCP server. Responses are structured to be useful alongside other MCP servers and tools. Note that if you're combining regulated data across systems, your own governance process may require a review of that combined data flow.
Does the MCP Server work with our SSO? Yes. Authorization redirects to your normal Risk Cloud login page, including SAML single sign-on.
Do I need an admin to set this up for me? Setup happens in two steps:
- Your administrator does the one-time setup. They enable the MCP Server for your environment and assign the MCP Access entitlement to your Risk Cloud user.
- Then you connect your own AI tool to LogicGate. This step must be done by every user who wants to use the MCP server, even after the admin setup is complete.
Note that your organization may separately restrict who can add connectors within the AI tool itself or, depending on your tools, may be able to publish custom connectors that make step 2 easier for your users.
Can I get reports and dashboards through MCP? Not in this release. The ability to reference dashboards, visual reports, and table reports is planned for a later phase of our MCP server.
Troubleshooting
| Symptom | What to check |
|---|---|
| Connection is refused, or you get a 401 | Confirm the MCP Server is enabled in your environment (contact your account team) and that your user has the MCP access entitlement. |
| 404 on the MCP address | Confirm the address is your environment URL plus /mcp, with no trailing slash, and that you used https. |
| Authorization never completes | Try signing in to Risk Cloud in a browser tab first, then retry the authorization from your AI tool. |
| "Add custom connector" is missing | Your organization may restrict custom connectors in your AI tool. Contact your IT or AI governance team. |
| Prompted to log in repeatedly | Expected after your session expires. If it happens more often than described above, contact support. |
| Results look incomplete | Tools return bounded result sets and signal truncation. Ask your AI tool to narrow the question (a specific workflow, a specific filter) rather than asking for everything. |
| The tool asks for an OAuth client ID and secret | Risk Cloud registers clients automatically and doesn't issue static credentials. If the tool has no automatic-registration option, it can't connect yet. Please tell your account team which tool it is so we can explore this. |
| You can't tell whether the problem is Risk Cloud or your AI tool | Test the endpoint with MCP Inspector (npx @modelcontextprotocol/inspector). If Inspector connects and your tool doesn't, the issue is on the tool side. |
| Your AI tool isn't in the verified list | It may still work — see Connecting other MCP clients. If it doesn't connect, contact your account team so we can verify that tool's behavior. |
If you run into issues that aren't covered here, contact support@logicgate.com.
Information for AI Governance Teams
Many organizations require an AI governance review before a new AI integration is approved. This section is written to be handed to that reviewer.
The single most important point
The Risk Cloud MCP Server does not perform AI inference and does not send your data to any model provider.
It is a read-only, permission-scoped API surface, packaged in a standard format that AI tools can consume. The large language model in this architecture lives entirely in the AI tool your organization chose and already governs. LogicGate does not select the model, does not see the prompts your users write, and does not transmit Risk Cloud data to OpenAI, Anthropic, or any other model provider as part of the MCP Server.
Practically, this means the AI risk assessment for this integration is mostly an assessment of the AI tool your organization is already using, plus a data-access review of what that tool can now read.
Data flow
| Stage | What happens |
|---|---|
| Authentication | The user authenticates to Risk Cloud through your existing login, including SAML SSO. OAuth 2.1 with PKCE; short-lived, narrowly scoped tokens. |
| Request | The user's AI tool calls a read tool on your Risk Cloud environment's MCP endpoint over HTTPS. |
| Authorization | Risk Cloud validates the token, resolves the user, and applies that user's entitlements. The tool's required read scope is checked at call time. |
| Response | Risk Cloud returns structured data, limited to what the user is permitted to see, over HTTPS to the user's AI tool. |
| Inference | Occurs entirely inside the customer's AI tool, under the customer's own agreement with that vendor. LogicGate is not a party to it. |
| Retention by LogicGate | LogicGate does not receive or retain your prompts. Your LLM takes the prompt and uses it to call the relevant MCP tools. |
| Retention by the AI tool | Governed by your agreement with that vendor. This is the retention question your governance team should focus on. |
Controls a reviewer will likely ask about
| Control | Status |
|---|---|
| Authentication standard | OAuth 2.1, authorization code flow with PKCE (S256 required; plain rejected). State parameter required. |
| Identity model | Named individual users only. No shared or service accounts in this release. |
| Least privilege | Read-only scopes, intersected with the user's existing Risk Cloud entitlements. Default deny. |
| Write / destructive actions | None available. Enforced by a curated tool list, a runtime read-only verb check, and per-tool scope authorization. |
| Tenant isolation | Single-tenant instance per environment. Cross-tenant access is not architecturally possible. |
| Session isolation | Each session is scoped to the authenticated user; no cross-session data sharing. |
| Transport security | HTTPS only. |
| Token lifetime | Access token 10 minutes; refresh token 24 hours. Per-client revocation supported. |
| Model training on your data | Not applicable to the MCP Server. No model is involved. Your AI tool's training terms apply to its own handling of the data. |
| Administrative control | The MCP Server can be disabled for the environment by an administrator. |
| Prompt injection / tool poisoning | Tool definitions are authored and reviewed by LogicGate and are not customer-editable. Because all tools are read-only and permission-scoped, a compromised prompt cannot cause a write, escalate privileges, or reach data the user cannot already see. |
Questions your governance team should answer internally
The MCP Server extends your existing Risk Cloud access into a tool you already run. The open questions are usually about that tool, not about Risk Cloud:
- Which AI tool will connect, and is it already approved?
- What is that tool's data retention and training policy for the data it receives?
- Which users should be permitted to connect, and do their Risk Cloud entitlements reflect what you're comfortable exposing to an AI tool?
- Are other MCP servers connected to the same AI tool, and does chaining Risk Cloud data with those systems create a new data flow that needs its own review?
- Does your network policy allow the tool to reach your Risk Cloud subdomain?
If your reviewer needs additional documentation, review the LogicGate Trust Center and, if you cannot find what you need, contact your LogicGate account team.