MCP Servers

The MCP Servers page shows every MCP server your agents connect to helping you identify unapproved servers, detect security problems, and control what tools agents are allowed to use.

What are MCP Servers

MCP stands for Model Context Protocol. An MCP server is a tool provider, a service that gives AI agents the ability to take specific actions, such as searching the web, reading a file, creating a Jira issue, or querying a database. When an agent needs to do something beyond generating text, it calls an MCP server.

Because MCP servers extend what agents can do, they are an important security surface. An unapproved or compromised MCP server can cause an agent to take actions your organization never intended.

This page helps you:
  • See which MCP servers are in use across your agent fleet.
  • Identify servers that have not been approved by your organization.
  • Detect posture problems such as dormant servers or missing authentication.
  • Control which servers agents are allowed to use.

MCP Summary

Three numbers in the page header give you an immediate picture of your MCP server environment:
Number What it means
Servers The total number of MCP servers Agent Identity Security has discovered that your agents connect to.
Unsanctioned Servers that have not been officially approved by your organization. These carry supply-chain risk because they may have been added by individual agents or developers without security review.
Posture Issues The total count of open security problems detected across all MCP servers.

Analysis panel

When you open the MCP Servers page, an Analysis panel appears above the full server list. It highlights where problems are concentrated so you know where to start.

Top 5 Risky MCP Servers

The five MCP servers with the most active posture issues, ranked in order. Each entry shows the server name, the number of issues, and a proportional red bar. Select a server name to go directly to its detail page. Select Show in table to filter the full list to servers with active issues.

MCP Security Findings

A list of security problem types found across your MCP servers, with a count of how many times each appears and how many agents are affected. Common finding types include:

Finding type What it means
Agent Using Unapproved MCP Server An agent is connecting to a server that is not on the approved list.
Local MCP Server Detected An MCP server is running locally on a device rather than a managed endpoint.
Dormant MCP Server A server has not been used within the configured inactivity threshold.
MCP Tool Parameter Mutation A server's tool definitions have changed since they were last approved.
MCP Authentication Missing A server has no authentication configured, meaning any agent can call it.

Select View in Agent Posture to see the full list of issues in context.

All Servers Inventory

Below the Analysis panel is the full list of all MCP servers. Search, sort, and filter by server name, status, posture issues, or sanction status.
Column What it shows
Server Name The server's name. A secondary line shows the agent that provided it.
Status Active means the server is reachable and in use. Inactive means it is not responding or has been disconnected.
Tools Count The number of actions (tools) this server exposes to agents.
Credentials Count The number of authentication credentials configured for this server.
Posture Issues The count of open security problems for this server.
Connected Agents The agents using this server. Hover the +N badge to see all agent names.
Sanctioned A toggle switch. Turn it on to mark the server as organization-approved. Turn it off to mark it as unsanctioned. Using an unsanctioned MCP server raises a posture issue for the agent. Once the MCP server is sanctioned, the posture issue is automatically resolved.

MCP Server Detail Page

Select a server row in the All Servers inventory to open the MCP Server Detail page for that server.

The detail page header contains the following fields:
Field Description
Server Name The display name of the MCP server (for example, "atlassian").
Server Type Label The text "MCP Server" beside the name, confirming the resource type.
URL The endpoint URL of the MCP server (for example, https://mcp.atlassian.com/v1/mcp/authv2). A copy icon beside the URL copies it to the clipboard.
Tools Badge Shows the number of tools exposed by this server (for example, "17 Tools").
Status Badge Colour-coded pill showing "Active" or "Inactive".
Sanctioned Badge Shows "Sanctioned" or "Unsanctioned" to indicate approval status.

MCP Server Details

Click any row in the All Servers grid navigates to the MCP Server Detail page.

The server detail page provides four tabs Posture Issues (displayed by default), Tools, Credentials, and Connected Agents.

Posture Issues Tab
Displays all open security and compliance issues detected for the MCP server in a data grid with search, column customization, density, refresh, and export controls.
Column Description
Posture Issue Name of the detected security or compliance violation (for example, "New MCP Server Attached" or "Agent Using Unapproved MCP Server"). Expandable rows reveal additional details.
Agent Name of the AI agent associated with this posture issue.
Resource The MCP server or resource affected by the issue.
Risk Level Severity rating: Critical, High, Medium, or Low. Displayed as a coloured badge with a matching severity icon.
Status Resolution state of the issue: Unresolved, In Progress, or Resolved.
Compliance Mapping Frameworks to which this issue maps, such as SOC 2, ISO/IEC 27001:2022, NIST SP 800-53 Rev. 5, and others. Additional frameworks beyond two are collapsed behind a +N chip that reveals the full list on hover.
Detected Date on which the issue was first detected (for example, "Jun 6, 2026").
Tools Tab
Lists all tools (MCP functions) exposed by the server. The subheader shows the total count (for example, "17 tools available on this server expand a row to see parameters"). The grid supports Search, Columns, and Density controls.
Column Description
Tool Name The programmatic name of the tool as declared by the MCP server (for example, getJiraIssue, createJiraIssue).
Description A plain-language summary of what the tool does. Long descriptions are truncated with a See more button that expands the full text inline.
Parameters The number of input parameters the tool accepts, displayed as a purple badge. Clicking the row expander (Expand row) reveals the parameter list with each parameter's name, data type, required flag, and description.
Tip: Use the row expander on a tool row to inspect its parameters before configuring a policy or reviewing a posture finding that references that specific tool.
Credentials Tab
Shows authentication credentials associated with this MCP server. The subheader displays the credential count (for example, "1 credential configured for authentication"). Each credential card shows:
Note: Each credential entry shows the Credential Name (e.g., "Authorization"), the Type of authentication method (e.g., Basic, OAuth, API Key), and the Expires date or "-" if no expiry is configured.
Connected Agents Tab
Connected Agents: Lists all AI agents currently using this MCP server, with a count badge on the tab label (e.g., Connected Agents (2)). Each agent is displayed as a card click any card to view the agent's full details.

Things to watch for

  • Any server showing Unsanctioned status needs a review before agents continue using it.
  • Dormant servers with no recent activity should be reviewed and potentially decommissioned.
  • Servers with zero credentials configured (MCP Authentication Missing) are open to any caller.
  • If a server's tool definitions have changed (MCP Tool Parameter Mutation), verify the change was intentional before allowing agents to continue calling it. Unexpected changes can indicate a supply-chain attack.