See which AI tools and MCP servers your people use
Oriant lists the AI clients installed on your machines and the MCP servers they are set up to use, and it records prompts sent on Claude web. Every call that passes through an MCP Connection is logged.
Installed AI clients on every machine
The Oriant endpoint agent scans each machine on a schedule and reports the AI clients it finds, such as Claude Desktop, Claude Code and Cursor. It finds them by app, config folder or running process. Each snapshot replaces the last one and shows on the machine’s page under Assets. Every report is tied to one person, through the email stamped on the machine at install, and Oriant never guesses identity from the machine. The inventory therefore tells you which person has which AI client.
MCP servers declared in AI client configs
The scan reads the MCP servers set up in each AI client’s config, including Claude Desktop Extensions, with the address each one connects to. Oriant compares them with your MCP Connections and sorts each server into one group.
- Through Oriant: the server points at an Oriant connection, so its calls are checked and recorded.
- Direct: the server connects straight to a remote server and skips Oriant.
- Local: the server runs on the machine.
- Unknown: the agent reported no address.
MCP servers that skip the gateway
The Coverage tab on the MCP Connections page counts the servers that do not go through Oriant and shows who has them. That list tells you whom to move behind the gateway. Oriant does not block a direct connection to an MCP server. It shows the connection on that machine.
AI tools and their sanction state
AI Tools lists every AI tool the agent has seen on your machines: who uses it, on how many machines, when it was first and last seen, and which AI clients are installed for it. Each tool has one sanction state. A newly found tool is Unsanctioned, which means its use is recorded and counted. You can set a tool to Sanctioned or to Blocked. The agent names the AI tool and the person behind AI traffic, and Blocked tools are rejected on the machine.
AI adoption from Claude web prompts
The agent records the prompts people submit on Claude web, together with the sender’s email. Responses are not recorded. Analytics turns the prompts into AI adoption by day, the busiest directory groups, AI tools by month and cohorts by each person’s first prompt. Prompt text is stored in your Oriant database, and Analytics shows counts, not prompt text. Use inside installed AI clients is not counted yet, and sorting prompts by kind of work is not turned on at app.oriant.app.
A log of every MCP call
Oriant records every MCP Call before it reaches the MCP server: the member, the connection, the tool, its arguments, the outcome and the timing. The outcome is succeeded, tool error, failed, refused or unknown. A refused call also shows the policy or argument rule that refused it, and a call that was sent but whose outcome could not be recorded is marked unknown and is never retried automatically. The tool’s result is not part of the record. Every member sees their own calls, and seeing other members’ calls needs a Permission that the Owner and Admin Roles hold.
Rolling out the endpoint agent
You can connect one machine with a single command to try it, or deploy to a fleet with a Microsoft Intune template. Every managed install needs a stamped email, so you configure that in Intune before you assign the template to devices. If the agent stops, the machine’s proxy settings are turned off so the machine keeps working.
Read the docs
More of Oriant
What can your AI reach today?
Oriant lists the AI clients, MCP servers, API keys and Skills on your machines, and it governs what AI does in your systems.