Agent Exposure Report: see what your fleet is actually running.
Find every AI agent, coding tool and MCP server on your endpoints, with the risks that matter flagged. One report for the engineers who fix things and the leaders who need the summary.
deny[msg] {
server := input.mcp.server
not server.pinned
msg := "unpinned MCP server"
}Sample report with illustrative data.
What an Agent Exposure Report looks like
This is a synthetic report, not a real fleet. It shows the inventory, MCP server configuration, hygiene flags and policy baseline you receive at the end of a read-only pilot, with every value labelled as illustrative.
Pilot cohort, illustrative data
24 endpoints, one engineering team. Report generated at the end of a 7 day read-only pilot.
AI agents and coding tools
Every agent and CLI found on the cohort, with the version in use.
| Tool | Version | Endpoints | Status |
|---|---|---|---|
| Claude Code | 2.1.4 | 18 | Approved |
| Cursor | 0.51.2 | 11 | Review |
| Codex CLI | 0.44.0 | 6 | Approved |
| GitHub Copilot | 1.302.0 | 22 | Approved |
| Local model runner | 0.9.1 | 2 | Unsanctioned |
MCP server configurations
Servers found in editor and CLI configuration, including ones added by hand.
| Server | Transport | Pinned | Endpoints | Flag |
|---|---|---|---|---|
| filesystem | stdio | Yes | 14 | Pinned |
| github | http | No | 9 | Unpinned |
| playwright | stdio | Yes | 6 | Pinned |
| warehouse-reader | stdio | No | 3 | Unpinned, plaintext token |
Hygiene and exposure flags
What the report escalates to the platform team, ordered by what needs action first.
- Down-level toolCursor 0.51.2 on 4 endpoints sits below the pinned baseline 0.52.1.
- Unpinned MCP server12 connections to 2 servers that are not pinned to a version.
- Credential in agent config1 configuration file holds a static token in plaintext for warehouse-reader.
- Unrecognized toolA local model runner on 2 endpoints is not on the approved list.
Policy baseline
How current usage compares with the default policy, evaluated on every MCP call.
The baseline requires pinned MCP servers. Over the 7 day window 23 MCP calls matched the rule. All 23 were logged and none were blocked, because the cohort ran in monitor mode. Turning the same rule into enforcement is a later, explicit decision.
Synthetic example. Every hostname, count and version on this page is illustrative, and the report contains no customer, tenant or real endpoint data.
Start a free pilotWhat the report covers
Every AI agent and coding tool
Claude Code, Cursor, Codex, Copilot and the ones nobody told you about, with versions.
Every MCP server configuration
Including servers added by hand to an editor on a laptop, which a network gateway never sees.
Hygiene and exposure flags
Out-of-date tools, unpinned or unverified MCP servers, plaintext credentials in agent configs.
A policy baseline
How current usage compares with a sensible default policy, in monitor mode, so nothing is blocked.
Scope and coverage
- Read-only
- Beacon is a user-space agent with no kernel driver. During the pilot it observes; it does not block or intercept anything.
- You choose the cohort
- Start with one team or a group of laptops, up to your whole fleet.
- What it covers
- AI agents, coding tools and MCP servers on the endpoints where Beacon runs. Machines without Beacon are not in the report.
- Easy to remove
- Uninstall Beacon at the end of the pilot if you decide not to continue.
How Alyria handles your data today, and what is planned, is on the security page.
How the free pilot works
- Step 1
Talk to us
A short call to agree the cohort and what you want to learn.
- Step 2
Deploy Beacon read-only
Install on the cohort with your usual software deployment tool.
- Step 3
Get your report
An inventory your platform team can act on, and a summary for leadership.