Granted Patent B1

US 12,568,091 · App. 19/267,475 · Granted Mar 3, 2026

Inventors: Andrew Morgan (Dublin, IE); Rahul Parwani (Atlanta, GA)

Assignee: Airia LLC

H04L63/105H04L63/102

Abstract

Claims (60)

1 . A method for securely executing artificial intelligence (“AI”) agents based on administrator-defined restrictions to model context protocol (“MCP”) servers, comprising:

receiving a resource profile that corresponds to a resource available at an MCP server, wherein the resource profile describes a tool and multiple resource commands;

causing display, at a user device, of a user interface (“UI”) for configuring AI agent usage of the resource, wherein the UI displays options for at least:

selecting restricted actions that correspond to at least one of a specific change to the resource profile, a change quantity of the resource profile, a restricted resource command to the MCP server, or a restricted output of the MCP server;

selecting triggers that correspond to the restricted actions, wherein each selected trigger defines a condition under which the corresponding restricted action causes a remedial action, wherein the condition comprises: a presence of the restricted action, a presence of a combination of the restricted action and at least one additional restricted action, or a weight of the restricted action towards a threshold; and

selecting the remedial actions that correspond to the triggers;

causing the UI to display selectable permissions that apply to the restricted actions, wherein the permissions comprise read and write permissions;

receiving UI selections from an administrative user, the UI selections comprising selected restricted actions, selected triggers, selected permissions, and selected remedial actions;

based on the received UI selections, creating a management profile that comprises the selected restricted actions, the selected permissions, the selected triggers, and the selected remedial actions; and

wherein an MCP orchestrator detects a first restricted action defined by the management profile, wherein the detection comprises:

identifying a first trigger defined in the management profile; and

triggering a first remedial action associated with the first trigger.

2 . The method of claim 1 , wherein detecting the first restricted action comprises identifying the specific change to the resource profile.

3 . The method of claim 1 , further comprising:

receiving UI selections to create first and second change thresholds; and

creating the first and second change thresholds, wherein the first change threshold corresponds to the first remedial action and the second change threshold corresponds to a second remedial action,

wherein detecting the first restricted action comprises detecting the change quantity by comparing a deviation amount of the resource profile against the first and second change thresholds.

4 . The method of claim 3 , wherein the deviation amount exceeds the first change threshold but not the second change threshold, triggering the first remedial action but not the second remedial action.

5 . The method of claim 3 , wherein the second change threshold represents a higher risk to data security than the first change threshold.

6 . The method of claim 5 , wherein the second remedial action comprises blocking a first resource command from reaching the MCP server, and wherein the first remedial action comprises displaying an alert.

7 . The method of claim 3 , wherein the first remedial action comprises applying the selected permissions of the management profile to a first resource command.

8 . The method of claim 1 , further comprising causing display, on the UI, of options for selecting allowed actions, wherein the allowed actions correspond to allowed resource commands and allowed changes to the resource profile, and wherein the allowed actions do not trigger the first remedial action.

9 . The method of claim 8 , wherein the selectable permissions are displayed for both the selected restricted actions and the selected allowed actions, and wherein selected permissions for a first allowed action are applied to a corresponding allowed resource command.

10 . The method of claim 1 , wherein the first remedial action comprises causing the UI to display an alert, and wherein displaying the alert comprises displaying selectable suggested functional remedial actions (“SFRAs”).

11 . The method of claim 10 , further comprising:

receiving selections of the SFRAs; and

causing execution of the SFRAs.

12 . The method of claim 11 , wherein causing execution of the SFRAs comprises terminating the MCP server and deleting a nano sandbox, wherein the MCP server executed within the nano sandbox.

13 . The method of claim 10 , the SFRAs are assigned to the first trigger based on the UI selections.

14 . The method of claim 1 , wherein the first restricted action corresponds to the restricted output from the MCP server, and wherein the remedial action comprises blocking the restricted output from being transmitted to an MCP client.

15 . The method of claim 14 , wherein the restricted output comprises a specific output having one or more keywords defined by administrator selections on the UI.

16 . The method of claim 14 , wherein the restricted output comprises a semantic meaning that is detected by an AI model.

17 . The method of claim 10 , wherein displaying the alert further comprises displaying multiple change thresholds and corresponding remedial actions, wherein the multiple change thresholds and corresponding remedial actions are defined in the management profile.

18 . The method of claim 17 , wherein corresponding remedial actions of the multiple change thresholds are selectable on the UI, causing execution of a selected one of the corresponding remedial actions.

19 . A non-transitory, computer-readable medium containing instructions for securely executing artificial intelligence (“AI”) agents based on administrator-defined restrictions to model context protocol (“MCP”) servers, wherein the instructions are executed by at least one processor to perform stages comprising:

receiving a resource profile that corresponds to a resource available at an MCP server, wherein the resource profile describes a tool and multiple resource commands;

causing display, at a user device, of a user interface (“UI”) for configuring AI agent usage of the resource, wherein the UI displays options for at least:

selecting restricted actions that correspond to at least one of a specific change to the resource profile, a change quantity of the resource profile, a restricted resource command to the MCP server, or a restricted output of the MCP server;

selecting triggers that correspond to the restricted actions, wherein each selected trigger defines a condition under which the corresponding restricted action causes a remedial action, wherein the condition comprises: a presence of the restricted action, a presence of a combination of the restricted action and at least one additional restricted action, or a weight of the restricted action towards a threshold; and

selecting the remedial actions that correspond to the triggers;

causing the UI to display selectable permissions that apply to the restricted actions, wherein the permissions comprise read and write permissions;

receiving UI selections from an administrative user, the UI selections comprising selected restricted actions, selected triggers, selected permissions, and selected remedial actions;

based on the received UI selections, creating a management profile that comprises the selected restricted actions, the selected permissions, the selected triggers, and the selected remedial actions; and

wherein an MCP orchestrator detects a first restricted action defined by the management profile, wherein the detection comprises:

identifying a first trigger defined in the management profile; and

triggering a first remedial action associated with the first trigger.

20 . A system for securely executing artificial intelligence (“AI”) agents based on administrator-defined restrictions to model context protocol (“MCP”) servers, the system comprising:

at least one physical non-transitory, computer-readable medium comprising instructions; and

at least one processor that executes the instructions to perform stages comprising:

receiving a resource profile that corresponds to a resource available at an MCP server, wherein the resource profile describes a tool and multiple resource commands;

causing display, at a user device, of a user interface (“UI”) for configuring AI agent usage of the resource, wherein the UI displays options for at least:

selecting restricted actions that correspond to at least one of a specific change to the resource profile, a change quantity of the resource profile, a restricted resource command to the MCP server, or a restricted output of the MCP server;

selecting triggers that correspond to the restricted actions, wherein each selected trigger defines a condition under which the corresponding restricted action causes a remedial action, wherein the condition comprises: a presence of the restricted action, a presence of a combination of the restricted action and at least one additional restricted action, or a weight of the restricted action towards a threshold; and

selecting the remedial actions that correspond to the triggers;

causing the UI to display selectable permissions that apply to the restricted actions, wherein the permissions comprise read and write permissions;

receiving UI selections from an administrative user, the UI selections comprising selected restricted actions, selected triggers, selected permissions, and selected remedial actions;

based on the received UI selections, creating a management profile that comprises the selected restricted actions, the selected permissions, the selected triggers, and the selected remedial actions; and

wherein an MCP orchestrator detects a first restricted action defined by the management profile, wherein the detection comprises:

identifying a first trigger defined in the management profile; and

triggering a first remedial action associated with the first trigger.

Assignments (1)

Continuity (1)

References Cited (22)

Cited By (2)

US 12,683,941 US 12,683,991