Use Tenable One OT Exposure via MCP Server
The Tenable One OT Exposure Model Context Protocol (MCP) server is a secure bridge between your AI assistants (for example, Claude or Copilot) and your Tenable One OT Exposure environment. Instead of navigating dashboards and menus, you can query assets, review telemetry, and retrieve operational technology (OT) data using natural language prompts.
With the MCP server, your AI client discovers and executes Tenable tools to query assets, evaluate risk scores, search vulnerabilities, and evaluate policies using your actual OT data.
The MCP server supports both primary deployment configurations:
-
A single ICP: Queries asset inventory, vulnerability search, active-operation management, and investigations for that site.
-
Enterprise Manager (multi-site manager): Executes queries across multiple connected sites and aggregates the results into a single response without requiring separate logins to each instance.
Prerequisites
Before you connect to the Tenable One OT Exposure MCP server, verify that you have:
-
An active instance of Tenable One OT Exposure ICP or an Enterprise Manager connected to one or more ICPs.
-
A user account with required permissions. The MCP server validates your existing role and zone permissions on every call. You can view only the data that your user profile is permitted to access in the Tenable One OT Exposure user interface.
-
An MCP-compatible AI client that supports Streamable HTTP transport (for example, Claude Code, Claude Desktop, or Cursor).
-
One of the following authentication credentials:
-
Login credentials for Tenable One OT Exposure
-
A valid Tenable One OT Exposure API key.
Tip: To generate or retrieve your API keys, navigate to Settings > System Configuration > API Keys. For more information, see Generate API Keys.
-
MCP Server Connection Details
The MCP server endpoint resides at the /mcp path on both Tenable One OT Exposure ICP and Enterprise Manager.
-
Server URL: https://<your-server-address>/mcp
-
Transport: Streamable HTTP
Authentication
| Method | Use when | How it works |
|---|---|---|
| OAuth with PKCE | Interactive clients, and users authenticating via LDAP, Active Directory (AD), SAML, or local accounts. | The MCP client redirects you to sign in through a browser session. After you sign in, the client receives an identity bearer token. No client secret is required. |
| API Key | Scripts, developer tools, and custom integrations | The MCP client sends your API key in the X-APIKeys header with every request. |
API key header format:
X-APIKeys: key=<your-api-key>
Connect Claude Code
To connect Claude Code with OAuth:
-
In your terminal, run the command for your deployment type. Replace the placeholder <YOUR_SERVER_ADDRESS> with your server address:
Tenable One OT Exposure ICP:
Copyclaude mcp add tenable-ot --transport http https://<YOUR_SERVER_ADDRESS>/mcpEnterprise Manager:
Copyclaude mcp add tenable-ot-em --transport http https://<YOUR_SERVER_ADDRESS>/mcp -
Start Claude Code and run /mcp.
-
Select tenable-ot or tenable-ot-em and complete the login prompt in your browser.
Connect Using an API Key
To authenticate with an API key, supply the X-APIKeys header when registering the server:
-
Enterprise Manager
Copyclaude mcp add tenable-ot-em --transport http https://<YOUR_SERVER_ADDRESS>/mcp --header "X-APIKeys: key=<YOUR_API_KEY>" -
Tenable One OT Exposure ICP:
Copyclaude mcp add tenable-ot --transport http https://<YOUR_SERVER_ADDRESS>/mcp --header "X-APIKeys: key=<YOUR_API_KEY>"
Claude Code discovers the available Tenable tools automatically. To verify the connection, ask Claude to list your sites or display system information.
Connect Generic MCP Clients
For any client that supports Streamable HTTP, add a new MCP server with these settings:
-
Transport type: Streamable HTTP
-
URL: https://<your_server_address>/mcp
-
Authentication: OAuth with PKCE, or the X-APIKeys header with your API key.
For more information about how to add MCP servers, see your MCP client's documentation.
Available MCP Tools
During session initialization, the /mcp client calls tools/list per session to discover the available tools:
| Tool | Description |
get_asset
|
Returns an asset profile, including device identification, firmware version, Purdue Model level, criticality score, ports, and zones. |
list_zones
|
Lists all defined zones and their asset counts, or displays which zones an asset belongs to. |
list_sensors
|
Returns sensor operational status, traffic, and version. |
list_agents
|
Returns OT agent status, check-in timestamps, coverage, and scan activity. |
search_findings
|
Searches detection events by time, text, and filter. |
list_policies
|
Lists detection policies, their enabled state, and noise. |
get_system_info
|
Returns the server software version and mode. |
get_log_level
|
Returns service log levels. Requires an account with the Administrator role. |
list_sites
|
(Enterprise Manager only) Lists connected sites, operational health status, versions, license, and last synchronization timestamps. |
| search_vulnerabilities |
Searches known vulnerabilities by Common Vulnerabilities and Exposures (CVE) ID and CISA Known Exploited Vulnerabilities (KEV) status, and lists affected assets. |
Enterprise Manager Behavior
On Enterprise Manager, single-site ICP tools are exposed with a sites_<tool> (for example, sites_search_assets). These tools run on the connected sites and return results per site.
-
Unfiltered Queries: The MCP server executes queries on all connected ICPs in parallel and returns an aggregated result per site (for example: "Show me all high-risk assets across all my sites”).
-
Filtered Queries: The MCP server limits execution to the specific sites identified in the prompt (for example, "List all Siemens PLCs on Site 1").
-
Site Failures: If an individual site goes offline or denies access, the MCP server returns valid results from the remaining sites and flags the affected site in the response payload.
-
Changes at Runtime: When an ICP connects or disconnects from Enterprise Manager, the tool list updates automatically without any manual configuration.
Audit Logging
Tenable One OT Exposure records all operations executed through the MCP server in the system audit log with source: mcp, so you can filter MCP activity separately from actions in the user interface or REST API.
| Field | Description |
source
|
Identifies the event origin as mcp. |
user_id
|
The authenticated user. |
auth_method
|
Shows whether api_key or oauth method is used. |
tool
|
The tool called for the process. |
sites
|
The sites targeted (Enterprise Manager only). |
llm_client / llm_client_version |
The MCP client name and version, for example claude-code. |
duration_ms
|
Total time for the call, including requests to multiple sites. |
success
|
Whether the call succeeded. On Enterprise Manager, true if at least one site returned results. |
error
|
The error message, including per-site errors on Enterprise Manager. |
Example Workflows
The following examples show questions you can ask your AI client after you connect. Your AI client decides which tools to call, and on Enterprise Manager it queries all connected sites unless you name specific ones.
Asset Inventory Queries
Ask your AI client:
- "Which Rockwell PLCs are present in the environment?"
- "How many Windows 10 hosts do I have?"
- "Are there any Schneider Electric devices with end-of-life status?"
The AI uses search_assets to filter by vendor, type, operating system, and end-of-life status, and get_asset to retrieve details such as firmware for individual assets.
Findings-Related Inquiries
Ask your AI client:
- "Am I affected by PHP2shell (CVE-2026-109618, CVE-2026-109619)?"
- "Do any assets contain exploitable Rockwell vulnerabilities?"
- "Are there any CVEs on the CISA KEV list affecting my servers or workstations?"
The AI uses search_vulnerabilities searches vulnerabilities by CVE, vendor, and known-exploited status, and then lists the affected assets.
Deployment-Related Queries
Ask your AI client:
- "What is the current health of my deployment and are any sensors degraded?"
- "Are there any groups of assets that are no longer being monitored?"
- "What additional sensors should I consider deploying? Agents? Zones?"
The AI uses list_sites (Enterprise Manager), list_sensors, list_agents, list_zones, and get_system_info to review site health, sensor status, agent coverage, and zone membership. For example, a disconnected sensor can explain why a site is degraded, and an agent that is not covering a network range can leave a group of assets unmonitored.
Permissions and Access Control
-
The MCP server validates user roles and zone assignments for every request.
-
Invocations inherit the privileges of the authenticated account.
-
A zone-restricted user sees only the assets and findings in their zones. For assets outside the user's zones, the server returns the "no asset matches, or it is outside your zones" message.
-
Analysts and Read-only users can execute the read tools. The get_log_level requires account assigned the Administrator role
Security, Governance, and Trust Architecture
| Control | Description |
|---|---|
| Human verification | The MCP server executes actions only in response to explicit user prompts. It does not perform automated, unsupervised tasks. |
| Controlled tool exposure | The server exposes only specific, designer-approved tools. It does not provide open-ended access to your systems. |
| RBAC and zone permissions | Existing RBAC and zone permissions are verified for every call. The MCP server does not add any new permission. |
| Authentication | Every call runs under the identity of the authenticated user. OAuth uses PKCE, which prevents the storage of static secrets on client endpoints. |
| Auditing | All MCP-originated actions are tagged source: mcp in the system log. |
| Data governance | Data processed by the external AI assistant remains subject to your organization's enterprise AI governance and vendor terms of service. |