MCP Endpoints

Concept

Combine tools from multiple MCP servers behind a single endpoint.

Control exactly which tools are available through each endpoint.

Manage who can use each endpoint.

All tool invocations are logged regardless of which MCP server they target.

Automatically redact sensitive data from requests and responses.

Handle OAuth and token-based authentication for connected servers.

How It Works

graph LR
    A[AI Client<br/>Claude, ChatGPT, etc.] --> B[MCP Endpoint<br/>Hanzo KMS]
    B --> C[MCP Server 1<br/>Notion]
    B --> D[MCP Server 2<br/>GitHub]

Guide to Creating an MCP Endpoint

Head to your Agent Sentinel project and select MCP Endpoints from the sidebar, then click Create Endpoint.

Enter the following details for your endpoint:

• Name: A friendly name to identify this endpoint (e.g., "Engineering Team Endpoint")
• Description (Optional): A description of the endpoint's purpose
• Connected Servers: A selection of the MCP servers to make available through this endpoint

After creating the endpoint, you'll be taken to the endpoint details page. Here you can configure which tools from each connected server are available through this endpoint.

For each connected MCP server, you'll see a list of available tools. Toggle tools on or off to control what AI clients can access.

By default, no tools are enabled. You must explicitly enable the tools you want to make available.

The endpoint details page displays the Endpoint URL. Copy this URL—you'll need it to configure your AI client.

Connecting AI Clients

Add the endpoint to your Claude MCP configuration:

1. Open Claude settings
2. Navigate to the MCP section
3. Add a new server with your Hanzo KMS endpoint URL
4. Click Connect

When connecting for the first time, Claude will open an authorization page where you grant access to the endpoint. You can configure:

• Access Duration: How long the AI client can use the endpoint (e.g., 30 days)

After authorization, Claude can use all enabled tools from your endpoint.

Any MCP-compatible AI client can connect to your endpoint using the endpoint URL.

The general process is:

1. Locate the MCP server configuration in your AI client
2. Add your Hanzo KMS endpoint URL as a new server
3. Complete the authorization flow when prompted

Refer to your AI client's documentation for specific configuration steps.

Personal Credentials Authentication

Authentication Flow

Feedback
Authentication Prompt: After authorizing access to the endpoint, users are shown a list of all MCP servers that require their personal credentials.

Feedback
Authenticate Each Server: Users must authenticate with each server in the list. The authentication method depends on how the server was configured:
• OAuth: Users are redirected to the service (e.g., GitHub, Notion) to authorize access
• Bearer Token: Users enter their personal access token directly

Feedback
Re-authentication: Users can update their credentials for servers they've already authenticated with by clicking the Re-authenticate button. This is useful when tokens expire or when users want to switch accounts.
Feedback
Complete All Authentications: Users must authenticate with all servers requiring personal credentials before they can proceed. The connection will only be established once all required authentications are complete.

Access Control with Permission Conditions

Example Use Cases

Feedback
Team-specific endpoints: Create a role that only allows access to endpoints matching engineering-*
Feedback
Environment separation: Restrict production endpoints (prod-*) to senior team members

Configuring Permission Conditions

Feedback
Navigate to Access Control > Roles
Feedback
Edit or create a role
Feedback
Under MCP Endpoints permissions, click Add Condition
Feedback
Select the Endpoint Name property
Feedback
Choose an operator (equal, not equal, glob match, or in)
Feedback
Enter the value(s) to match

Supported Operators

PII Filtering

Supported PII Types

Configuring PII Filters

Open your MCP endpoint and locate the Filters section in the left column.

Click the edit icon to open the PII filter configuration modal.

Configure the following options:

• Filter Requests: Enable to redact PII from requests sent to MCP servers
• Filter Responses: Enable to redact PII from responses returned to AI clients
• PII Detection: Select which types of PII to detect and redact

Click Save to apply your PII filter settings. Changes take effect immediately for new tool invocations.

FAQ

When PII filtering is enabled, Hanzo KMS scans request and/or response payloads using pattern matching to detect sensitive data. Detected PII is replaced with redaction placeholders (e.g., [REDACTED_EMAIL]) before the data is passed through. The original data is never stored or logged when filtering is enabled.

Yes, you can enable PII filtering independently for requests (data sent to MCP servers) and responses (data returned to AI clients). This gives you granular control over where redaction is applied.

Hanzo KMS currently supports detection and redaction of: email addresses, phone numbers (US format), Social Security Numbers, credit card numbers, and IP addresses (both IPv4 and IPv6).

No. The selected PII types apply to both request and response filtering. If you need different filtering rules, consider creating separate endpoints.

Yes, you can connect an MCP server to as many endpoints as needed. Each endpoint can have different tools enabled, allowing you to create different access profiles.

When you disable a tool, AI clients connected to the endpoint will no longer be able to use it. The tool won't appear in the client's available tools list.

When an AI client connects to an endpoint, the user chooses an access duration (e.g., 30 days). After this period, the client will need to re-authorize.

Yes, you can revoke access by managing authorized sessions in the endpoint settings. This immediately disconnects the AI client.

If a connected MCP server uses personal credentials, users will be prompted to authenticate when they first connect to the endpoint. The authentication method depends on how the server was configured:

• OAuth servers: Users complete an OAuth authorization flow
• Bearer Token servers: Users enter their personal access token directly

This is a one-time process per server, and users can re-authenticate at any time if needed.